To see the other types of publications on this topic, follow the link: Generative Artificial Intelligence (GenAI).
Dissertations / Theses on the topic 'Generative Artificial Intelligence (GenAI)'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 50 dissertations / theses for your research on the topic 'Generative Artificial Intelligence (GenAI).'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
1
Benedetti, Riccardo. "From Artificial Intelligence to Artificial Art: Deep Learning with Generative Adversarial Networks." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2019. http://amslaurea.unibo.it/18167/.
Full textAbstract:
Neural Network had a great impact on Artificial Intelligence and nowadays the Deep Learning algorithms are widely used to extract knowledge from huge amount of data.
This thesis aims to revisit the evolution of Deep Learning from the origins till the current state-of-art by focusing on a particular prospective. The main question we try to answer is: can AI exhibit artistic abilities comparable to the human ones?
Recovering the definition of the Turing Test, we propose a similar formulation of the concept, indeed, we would like to test the machine's ability to exhibit artistic behaviour equivalent to, or indistinguishable from, that of a human.
The argument we will analyze as a support for this debate is an interesting and innovative idea coming from the field of Deep Learning, known as Generative Adversarial Network (GAN).
GAN is basically a system composed of two neural network fighting each other in a zero-sum game. The ''bullets'' fired during this challenge are simply images generated by one of the two networks.
The interesting part in this scenario is that, with a proper system design and training, after several iteration these fake generated images start to become more and more closer to the ones we see in the reality, making indistinguishable what is real from what is not.
We will talk about some real anecdotes around GANs to spice up even more the discussion generated by the question previously posed and we will present some recent real world application based on GANs to emphasize their importance also in term of business.
We will conclude with a practical experiment over an Amazon catalogue of clothing images and reviews with the aim of generating new never seen product starting from the most popular existing ones.
2
ABUKMEIL, MOHANAD. "UNSUPERVISED GENERATIVE MODELS FOR DATA ANALYSIS AND EXPLAINABLE ARTIFICIAL INTELLIGENCE." Doctoral thesis, Università degli Studi di Milano, 2022. http://hdl.handle.net/2434/889159.
Full textAbstract:
For more than a century, the methods of learning representation and the exploration of the intrinsic structures of data have developed remarkably and currently include supervised, semi-supervised, and unsupervised methods. However, recent years have witnessed the flourishing of big data, where typical dataset dimensions are high, and the data can come in messy, missing, incomplete, unlabeled, or corrupted forms. Consequently, discovering and learning the hidden structure buried inside such data becomes highly challenging.
From this perspective, latent data analysis and dimensionality reduction play a substantial role in decomposing the exploratory factors and learning the hidden structures of data, which encompasses the significant features that characterize the categories and trends among data samples in an ordered manner. That is by extracting patterns, differentiating trends, and testing hypotheses to identify anomalies, learning compact knowledge, and performing many different machine learning (ML) tasks such as classification, detection, and prediction.
Unsupervised generative learning (UGL) methods are a class of ML characterized by their possibility of analyzing and decomposing latent data, reducing dimensionality, visualizing the manifold of data, and learning representations with limited levels of predefined labels and prior assumptions. Furthermore, explainable artificial intelligence (XAI) is an emerging field of ML that deals with explaining the decisions and behaviors of learned models. XAI is also associated with UGL models to explain the hidden structure of data, and to explain the learned representations of ML models. However, the current UGL models lack large-scale generalizability and explainability in the testing stage, which leads to restricting their potential in ML and XAI applications.
To overcome the aforementioned limitations, this thesis proposes innovative methods that integrate UGL and XAI to enable data factorization and dimensionality reduction to improve the generalizability of the learned ML models. Moreover, the proposed methods enable visual explainability in modern applications as anomaly detection and autonomous driving systems. The main research contributions are listed as follows:
• A novel overview of UGL models including blind source separation (BSS), manifold learning (MfL), and neural networks (NNs). Also, the overview considers open issues and challenges among each UGL method.
• An innovative method to identify the dimensions of the compact feature space via a generalized rank in the application of image dimensionality reduction.
• An innovative method to hierarchically reduce and visualize the manifold of data to improve the generalizability in limited data learning scenarios, and computational complexity reduction applications.
• An original method to visually explain autoencoders by reconstructing an attention map in the application of anomaly detection and explainable autonomous driving systems.
The novel methods introduced in this thesis are benchmarked on publicly available datasets, and they outperformed the state-of-the-art methods considering different evaluation metrics. Furthermore, superior results were obtained with respect to the state-of-the-art to confirm the feasibility of the proposed methodologies concerning the computational complexity, availability of learning data, model explainability, and high data reconstruction accuracy.
3
Konuko, Goluck. "Low Bitrate Face Video Compression with Generative Animation Models." Electronic Thesis or Diss., université Paris-Saclay, 2025. http://www.theses.fr/2025UPAST014.
Full textAbstract:
Cette thèse aborde le défi de réaliser une compression vidéo à ultra-faible débit pour la vidéoconférence, en se concentrant sur la préservation d'une haute qualité visuelle tout en minimisant la bande passante de transmission. Les codecs traditionnels comme HEVC et VVC éprouvent des difficultés à très bas débits, particulièrement pour représenter avec précision les expressions faciales dynamiques, les mouvements de tête et les occlusions, qui sont essentiels pour le réalisme et la précision dans la communication en face à face. Pour surmonter ces limitations, cette recherche développe une méthode de compression basée sur l'apprentissage utilisant des modèles génératifs profonds et améliore leurs performances grâce à la transmission d'informations auxiliaires ou au codage prédictif à bas débit.Le Deep Animation Codec (DAC) est introduit comme une solution qui utilise des modèles génératifs pour encoder les mouvements faciaux liés à la parole via une représentation compacte des points clés de mouvement, réalisant ainsi des réductions substantielles du débit binaire. Pour traiter les limitations du DAC avec des poses de tête complexes et des occlusions, le Multi-Reference DAC (MRDAC) utilise plusieurs images de référence et l'apprentissage contrastif pour améliorer la précision de reconstruction dans des conditions difficiles. En s'appuyant sur cela, le Hybrid Deep Animation Codec (HDAC) intègre des codecs vidéo traditionnels avec des cadres génératifs pour atteindre une qualité adaptative, encore améliorée par l'apprentissage à débit binaire variable et un mécanisme de transfert des hautes fréquences (HF) pour une reconstruction détaillée. Enfin, nous avons exploré une approche de codage prédictif à ultra-faible débit, mettant en évidence les défis associés et les outils d'optimisation qui peuvent être utilisés pour apprendre efficacement un codage résiduel compact à bas débit. Plus précisément, le cadre de codage prédictif proposé (RDAC) exploite les dépendances temporelles et l'apprentissage résiduel conditionnel pour atteindre un compromis robuste entre la perte d'information et l'évolutivité de la qualité dans les contraintes du codage à faible débit.Collectivement, ces contributions font progresser le domaine en permettant une compression vidéo robuste et de haute qualité à ultra-faible débit, améliorant la faisabilité de la vidéoconférence et les applications potentielles en réalité virtuelle ainsi que le stockage efficace de contenu vidéo de type « talking-head »<br>This thesis addresses the challenge of achieving ultra-low bitrate video compression for video conferencing by focusing on the preservation of high visual quality while minimizing transmission bandwidth. Traditional codecs like HEVC and VVC struggle at very low bitrates, particularly with accurately representing dynamic facial expressions, head movements, and occlusions, which are important for realism and accuracy in face-to-face communication. To overcome these limitations, this research develops learning-based compression method using deep generative models and enhances their performance through side information transmission or predictive coding at low bitrates.The Deep Animation Codec (DAC) is introduced as a solution that uses generative models to encode speech-related facial motion through a compact representation of motion keypoints, achieving substantial bitrate reductions. To address DAC's limitations with complex head poses and occlusions, the Multi-Reference DAC (MRDAC) uses multiple reference frames and contrastive learning to enhance reconstruction accuracy under challenging conditions. Building on this, the Hybrid Deep Animation Codec (HDAC) integrates traditional video codecs with generative frameworks to achieve adaptive quality, further improved by variable bitrate learning and a High-Frequency (HF) shuttling mechanism for detailed reconstruction. Finally, we explored an approach to predictive coding at ultra-low bitrates showing the associated challenges and optimization tools that can be used to effectively learn compact residual coding at low bitrates. Specifically, the proposed predictive coding framework (RDAC) exploits temporal dependencies and conditional residual learning to achieve a robust trade-off between information loss and quality scalability within the constraints of low bitrate coding. Collectively, these contributions advance the field by enabling robust, high-quality video compression at ultra-low bitrates, enhancing the feasibility of video conferencing and potential applications in virtual reality and efficient storage of talking-head video content
4
Griffith, Todd W. "A computational theory of generative modeling in scientific reasoning." Diss., Georgia Institute of Technology, 1999. http://hdl.handle.net/1853/8177.
Full text
5
Navaroli, Nicholas Martin. "Generative Probabilistic Models for Analysis of Communication Event Data with Applications to Email Behavior." Thesis, University of California, Irvine, 2015. http://pqdtopen.proquest.com/#viewpdf?dispub=3668831.
Full textAbstract:
<p> Our daily lives increasingly involve interactions with others via different communication channels, such as email, text messaging, and social media. In this context, the ability to analyze and understand our communication patterns is becoming increasingly important. This dissertation focuses on generative probabilistic models for describing different characteristics of communication behavior, focusing primarily on email communication. </p><p> First, we present a two-parameter kernel density estimator for estimating the probability density over recipients of an email (or, more generally, items which appear in an itemset). A stochastic gradient method is proposed for efficiently inferring the kernel parameters given a continuous stream of data. Next, we apply the kernel model and the Bernoulli mixture model to two important prediction tasks: given a partially completed email recipient list, 1) predict which others will be included in the email, and 2) rank potential recipients based on their likelihood to be added to the email. Such predictions are useful in suggesting future actions to the user (i.e. which person to add to an email) based on their previous actions. We then investigate a piecewise-constant Poisson process model for describing the time-varying communication rate between an individual and several groups of their contacts, where changes in the Poisson rate are modeled as latent state changes within a hidden Markov model. </p><p> We next focus on the time it takes for an individual to respond to an event, such as receiving an email. We show that this response time depends heavily on the individual's typical daily and weekly patterns - patterns not adequately captured in standard models of response time (e.g. the Gamma distribution or Hawkes processes). A time-warping mechanism is introduced where the absolute response time is modeled as a transformation of effective response time, relative to the daily and weekly patterns of the individual. The usefulness of applying the time-warping mechanism to standard models of response time, both in terms of log-likelihood and accuracy in predicting which events will be quickly responded to, is illustrated over several individual email histories.</p>
6
Franceschelli, Giorgio. "Generative Deep Learning and Creativity." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2021.
Find full textAbstract:
“Non ha la presunzione di originare nulla; può solo fare ciò che noi sappiamo ordinarle di fare”. Così, oltre 150 anni fa, Lady Lovelace commentava la Macchina Analitica di Babbage, l’antenato dei nostri computer. Una frase che, a distanza di tanti anni, suona quasi come una sfida: grazie alla diffusione delle tecniche di Generative Deep Learning e alle ricerche nell’ambito della Computational Creativity, sempre più sforzi sono stati destinati allo smentire l’ormai celebre Obiezione della Lovelace. Proprio a partire da questa, quattro domande formano i capisaldi della Computational Creativity: se è possibile sfruttare tecniche computazionali per comprendere la creatività umana; e, soprattutto, se i computer possono fare cose che sembrino creative (se non che siano effettivamente creative), e se possono imparare a riconoscere la creatività. Questa tesi si propone dunque di inserirsi in tale contesto, esplorando queste ultime due questioni grazie a tecniche di Deep Learning.
In particolare, sulla base della definizione di creatività proposta da Margaret Boden, è presentata una metrica data dalla somma pesata di tre singole componenti (valore, novità e sorpresa) per il riconoscimento della creatività. In aggiunta, sfruttando tale misura, è presentato anche UCAN (Unexpectedly Creative Adversarial Network), un modello generativo orientato alla creatività, che impara a produrre opere creative massimizzando la metrica di cui sopra. Sia il generatore sia la metrica sono stati testati sul contesto della poesia americana del diciannovesimo secolo; i risultati ottenuti mostrano come la metrica sia effettivamente in grado di intercettare la traiettoria storica, e come possa rappresentare un importante passo avanti per lo studio della Computational Creativity; il generatore, pur non ottenendo risultati altrettanto eccellenti, si pone quale punto di partenza per la definizione futura di un modello effettivamente creativo.
7
Nilsson, Alexander, and Martin Thönners. "A Framework for Generative Product Design Powered by Deep Learning and Artificial Intelligence : Applied on Everyday Products." Thesis, Linköpings universitet, Maskinkonstruktion, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-149454.
Full textAbstract:
In this master’s thesis we explore the idea of using artificial intelligence in the product design process and seek to develop a conceptual framework for how it can be incorporated to make user customized products more accessible and affordable for everyone. We show how generative deep learning models such as Variational Auto Encoders and Generative Adversarial Networks can be implemented to generate design variations of windows and clarify the general implementation process along with insights from recent research in the field. The proposed framework consists of three parts: (1) A morphological matrix connecting several identified possibilities of implementation to specific parts of the product design process. (2) A general step-by-step process on how to incorporate generative deep learning. (3) A description of common challenges, strategies andsolutions related to the implementation process. Together with the framework we also provide a system for automatic gathering and cleaning of image data as well as a dataset containing 4564 images of windows in a front view perspective.
8
Delacruz, Gian P. "Using Generative Adversarial Networks to Classify Structural Damage Caused by Earthquakes." DigitalCommons@CalPoly, 2020. https://digitalcommons.calpoly.edu/theses/2158.
Full textAbstract:
The amount of structural damage image data produced in the aftermath of an earthquake can be staggering. It is challenging for a few human volunteers to efficiently filter and tag these images with meaningful damage information. There are several solution to automate post-earthquake reconnaissance image tagging using Machine Learning (ML) solutions to classify each occurrence of damage per building material and structural member type. ML algorithms are data driven; improving with increased training data. Thanks to the vast amount of data available and advances in computer architectures, ML and in particular Deep Learning (DL) has become one of the most popular image classification algorithms producing results comparable to and in some cases superior to human experts. These kind of algorithms need the input images used for the training to be labeled, and even if there is a large amount of images most of them are not labeled and it takes structural engineers a large amount of time to do it. The current data earthquakes image data bases do not contain the label information or is incomplete slowing significantly the advance of a solution and are incredible difficult to search. To be able to train a ML algorithm to classify one of the structural damages it took the architecture school an entire year to gather 200 images of the specific damage. That number is clearly not enough to avoid overfitting so for this thesis we decided to generate synthetic images for the specific structural damage. In particular we attempt to use Generative Adversarial Neural Networks (GANs) to generate the synthetic images and enable the fast classification of rail and road damage caused by earthquakes. Fast classification of rail and road damage can allow for the safety of people and to better prepare the reconnaissance teams that manage recovery tasks. GANs combine classification neural networks with generative neural networks. For this thesis we will be combining a convolutional neural network (CNN) with a generative neural network. By taking a classifier trained in a GAN and modifying it to classify other images the classifier can take advantage of the GAN training without having to find more training data. The classifier trained in this way was able to achieve an 88\% accuracy score when classifying images of structural damage caused by earthquakes.
9
Mallik, Mohammed Tariqul Hassan. "Electromagnetic Field Exposure Reconstruction by Artificial Intelligence." Electronic Thesis or Diss., Université de Lille (2022-....), 2023. https://pepite-depot.univ-lille.fr/ToutIDP/EDENGSYS/2023/2023ULILN052.pdf.
Full textAbstract:
Le sujet de l'exposition aux champs électromagnétiques a fait l'objetd'une grande attention à la lumière du déploiement actuel du réseaucellulaire de cinquième génération (5G). Malgré cela, il reste difficilede reconstituer avec précision le champ électromagnétique dans unerégion donnée, faute de données suffisantes. Les mesures in situ sontd'un grand intérêt, mais leur viabilité est limitée, ce qui renddifficile la compréhension complète de la dynamique du champ. Malgré legrand intérêt des mesures localisées, il existe encore des régions nontestées qui les empêchent de fournir une carte d'exposition complète. Larecherche a exploré des stratégies de reconstruction à partird'observations provenant de certains sites localisés ou de capteursdistribués dans l'espace, en utilisant des techniques basées sur lagéostatistique et les processus gaussiens. En particulier, desinitiatives récentes se sont concentrées sur l'utilisation del'apprentissage automatique et de l'intelligence artificielle à cettefin. Pour surmonter ces problèmes, ce travail propose de nouvellesméthodologies pour reconstruire les cartes d'exposition aux CEM dans unezone urbaine spécifique en France. L'objectif principal est dereconstruire des cartes d'exposition aux ondes électromagnétiques àpartir de données provenant de capteurs répartis dans l'espace. Nousavons proposé deux méthodologies basées sur l'apprentissage automatiquepour estimer l'exposition aux ondes électromagnétiques. Pour la premièreméthode, le problème de reconstruction de l'exposition est défini commeune tâche de traduction d'image à image. Tout d'abord, les données ducapteur sont converties en une image et l'image de référencecorrespondante est générée à l'aide d'un simulateur basé sur le tracédes rayons. Nous avons proposé un réseau adversarial cGAN conditionnépar la topologie de l'environnement pour estimer les cartes d'expositionà l'aide de ces images. Le modèle est entraîné sur des images de cartesde capteurs tandis qu'un environnement est donné comme entréeconditionnelle au modèle cGAN. En outre, la cartographie du champélectromagnétique basée sur le Generative Adversarial Network estcomparée au simple Krigeage. Les résultats montrent que la méthodeproposée produit des estimations précises et constitue une solutionprometteuse pour la reconstruction des cartes d'exposition. Cependant,la production de données de référence est une tâche complexe car elleimplique la prise en compte du nombre de stations de base actives dedifférentes technologies et opérateurs, dont la configuration du réseauest inconnue, par exemple les puissances et les faisceaux utilisés parles stations de base. En outre, l'évaluation de ces cartes nécessite dutemps et de l'expertise. Pour répondre à ces questions, nous avonsdéfini le problème comme une tâche d'imputation de données manquantes.La méthode que nous proposons prend en compte l'entraînement d'un réseauneuronal infini pour estimer l'exposition aux champs électromagnétiques.Il s'agit d'une solution prometteuse pour la reconstruction des cartesd'exposition, qui ne nécessite pas de grands ensembles d'apprentissage.La méthode proposée est comparée à d'autres approches d'apprentissageautomatique basées sur les réseaux UNet et les réseaux adversairesgénératifs conditionnels, avec des résultats compétitifs<br>The topic of exposure to electromagnetic fields has received muchattention in light of the current deployment of the fifth generation(5G) cellular network. Despite this, accurately reconstructing theelectromagnetic field across a region remains difficult due to a lack ofsufficient data. In situ measurements are of great interest, but theirviability is limited, making it difficult to fully understand the fielddynamics. Despite the great interest in localized measurements, thereare still untested regions that prevent them from providing a completeexposure map. The research explored reconstruction strategies fromobservations from certain localized sites or sensors distributed inspace, using techniques based on geostatistics and Gaussian processes.In particular, recent initiatives have focused on the use of machinelearning and artificial intelligence for this purpose. To overcome theseproblems, this work proposes new methodologies to reconstruct EMFexposure maps in a specific urban area in France. The main objective isto reconstruct exposure maps to electromagnetic waves from some datafrom sensors distributed in space. We proposed two methodologies basedon machine learning to estimate exposure to electromagnetic waves. Forthe first method, the exposure reconstruction problem is defined as animage-to-image translation task. First, the sensor data is convertedinto an image and the corresponding reference image is generated using aray tracing-based simulator. We proposed an adversarial network cGANconditioned by the environment topology to estimate exposure maps usingthese images. The model is trained on sensor map images while anenvironment is given as conditional input to the cGAN model.Furthermore, electromagnetic field mapping based on the GenerativeAdversarial Network is compared to simple Kriging. The results show thatthe proposed method produces accurate estimates and is a promisingsolution for exposure map reconstruction. However, producing referencedata is a complex task as it involves taking into account the number ofactive base stations of different technologies and operators, whosenetwork configuration is unknown, e.g. powers and beams used by basestations. Additionally, evaluating these maps requires time andexpertise. To answer these questions, we defined the problem as amissing data imputation task. The method we propose takes into accountthe training of an infinite neural network to estimate exposure toelectromagnetic fields. This is a promising solution for exposure mapreconstruction, which does not require large training sets. The proposedmethod is compared with other machine learning approaches based on UNetnetworks and conditional generative adversarial networks withcompetitive results
10
Busatta, Gianluca. "Italian Retrieval-Augmented Generative Question Answering System for Legal Domains." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2022.
Find full textAbstract:
A typical scenario involves a user searching an information about something and obtaining a list of documents from an information retrieval system. The retrieved documents may be more or less relevant and it could happen that the information sought is contained in several documents. This would possibly leave the task of searching the information in different documents to the user. In this thesis, it is has been developed an Italian question answering system for legal domains with a Retrieval-Augmented Generation (RAG) approach that aims to directly satisfy the information need of the user. The model is composed of a retriever and a generator both of which are based on Transformer and it has been trained firstly in a self-supervised way on the library of Gruppo Maggioli company, and then in a supervised way on a novel Italian question answering dataset build on purpose. Once the user has provided an input, the model automatically retrieves possibly relevant documents from the knowledge base and use them to condition the generation of an appropriate answer.
11
Misino, Eleonora. "Deep Generative Models with Probabilistic Logic Priors." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2021. http://amslaurea.unibo.it/24058/.
Full textAbstract:
Many different extensions of the VAE framework have been introduced in the past. How ever, the vast majority of them focused on pure sub-symbolic approaches that are not sufficient for solving generative tasks that require a form of reasoning. In this thesis, we propose the probabilistic logic VAE (PLVAE), a neuro-symbolic deep generative model that combines the representational power of VAEs with the reasoning ability of probabilistic logic programming. The strength of PLVAE resides in its probabilistic logic prior, which provides an interpretable structure to the latent space that can be easily changed in order to apply the model to different scenarios. We provide empirical results of our approach by training PLVAE on a base task and then using the same model to generalize to novel tasks that involve reasoning with the same set of symbols.
12
Kalantari, John I. "A general purpose artificial intelligence framework for the analysis of complex biological systems." Diss., University of Iowa, 2017. https://ir.uiowa.edu/etd/5953.
Full textAbstract:
This thesis encompasses research on Artificial Intelligence in support of automating scientific discovery in the fields of biology and medicine. At the core of this research is the ongoing development of a general-purpose artificial intelligence framework emulating various facets of human-level intelligence necessary for building cross-domain knowledge that may lead to new insights and discoveries. To learn and build models in a data-driven manner, we develop a general-purpose learning framework called Syntactic Nonparametric Analysis of Complex Systems (SYNACX), which uses tools from Bayesian nonparametric inference to learn the statistical and syntactic properties of biological phenomena from sequence data. We show that the models learned by SYNACX offer performance comparable to that of standard neural network architectures. For complex biological systems or processes consisting of several heterogeneous components with spatio-temporal interdependencies across multiple scales, learning frameworks like SYNACX can become unwieldy due to the the resultant combinatorial complexity. Thus we also investigate ways to robustly reduce data dimensionality by introducing a new data abstraction. In particular, we extend traditional string and graph grammars in a new modeling formalism which we call Simplicial Grammar. This formalism integrates the topological properties of the simplicial complex with the expressive power of stochastic grammars in a computation abstraction with which we can decompose complex system behavior, into a finite set of modular grammar rules which parsimoniously describe the spatial/temporal structure and dynamics of patterns inferred from sequence data.
13
Goodman, Genghis. "A Machine Learning Approach to Artificial Floorplan Generation." UKnowledge, 2019. https://uknowledge.uky.edu/cs_etds/89.
Full textAbstract:
The process of designing a floorplan is highly iterative and requires extensive human labor. Currently, there are a number of computer programs that aid humans in floorplan design. These programs, however, are limited in their inability to fully automate the creative process. Such automation would allow a professional to quickly generate many possible floorplan solutions, greatly expediting the process. However, automating this creative process is very difficult because of the many implicit and explicit rules a model must learn in order create viable floorplans. In this paper, we propose a method of floorplan generation using two machine learning models: a sequential model that generates rooms within the floorplan, and a graph-based model that finds adjacencies between generated rooms. Each of these models can be altered such that they are each capable of producing a floorplan independently; however, we find that the combination of these models outperforms each of its pieces, as well as a statistic-based approach.
14
Chow, Ka Nin. "An embodied cognition approach to the analysis and design of generative and interactive animation." Diss., Georgia Institute of Technology, 2010. http://hdl.handle.net/1853/34695.
Full textAbstract:
Animation is popularly thought of as a sequence of still images or cartoons that produce an illusion of movement. However, a broader perspective of animation should encompass the diverse kinds of media artifacts imbued with the illusion of life. In many multimedia artifacts today, computational media algorithmically implement expanded illusions of life, which include images not only moving, but also showing reactions to stimuli (reactive animation), transforming according to their own internal rules (autonomous animation), evolving over a period of time (metamorphic animation), or even generating varying instances subject to user intervention or chance (contingent animation).
Animation in the digital age consists of forms as varied as computer-generated imagery (CGI) in films, motion graphics on interactive multimedia websites, animated contents of video games, graphical interfaces of computer systems, and even digital signage in communal areas. With these forms, the new animation phenomena emerge from entertainment media, functional designs, and expressive works alike, all of which may engage viewers' sensory perceptions, cognitive processes, as well as motor actions.
Hence, the study and creation of animation now requires an interdisciplinary framework, including (1) insights from perceptual psychology and animation studies about animacy, (2) theories of conceptual blending from cognitive science applied to understanding images, (3) notions of embodiment and temporality in phenomenological approaches to human-computer interaction (HCI), and (4) new interpretations of liveness in performance studies accounts of computer-mediated performance. These emergent ideas jointly characterize the new role of animation in media, and produce new possibilities for more embodied, evocative, and affective forms of generative and interactive animation.
15
Adhikari, Aakriti. "Skin Cancer Detection using Generative Adversarial Networkand an Ensemble of deep Convolutional Neural Networks." University of Toledo / OhioLINK, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=toledo1574383625473665.
Full text
16
Kalchbrenner, Nal. "Encoder-decoder neural networks." Thesis, University of Oxford, 2017. http://ora.ox.ac.uk/objects/uuid:d56e48db-008b-4814-bd82-a5d612000de9.
Full textAbstract:
This thesis introduces the concept of an encoder-decoder neural network and develops architectures for the construction of such networks. Encoder-decoder neural networks are probabilistic conditional generative models of high-dimensional structured items such as natural language utterances and natural images. Encoder-decoder neural networks estimate a probability distribution over structured items belonging to a target set conditioned on structured items belonging to a source set. The distribution over structured items is factorized into a product of tractable conditional distributions over individual elements that compose the items. The networks estimate these conditional factors explicitly. We develop encoder-decoder neural networks for core tasks in natural language processing and natural image and video modelling. In Part I, we tackle the problem of sentence modelling and develop deep convolutional encoders to classify sentences; we extend these encoders to models of discourse. In Part II, we go beyond encoders to study the longstanding problem of translating from one human language to another. We lay the foundations of neural machine translation, a novel approach that views the entire translation process as a single encoder-decoder neural network. We propose a beam search procedure to search over the outputs of the decoder to produce a likely translation in the target language. Besides known recurrent decoders, we also propose a decoder architecture based solely on convolutional layers. Since the publication of these new foundations for machine translation in 2013, encoder-decoder translation models have been richly developed and have displaced traditional translation systems both in academic research and in large-scale industrial deployment. In services such as Google Translate these models process in the order of a billion translation queries a day. In Part III, we shift from the linguistic domain to the visual one to study distributions over natural images and videos. We describe two- and three- dimensional recurrent and convolutional decoder architectures and address the longstanding problem of learning a tractable distribution over high-dimensional natural images and videos, where the likely samples from the distribution are visually coherent. The empirical validation of encoder-decoder neural networks as state-of- the-art models of tasks ranging from machine translation to video prediction has a two-fold significance. On the one hand, it validates the notions of assigning probabilities to sentences or images and of learning a distribution over a natural language or a domain of natural images; it shows that a probabilistic principle of compositionality, whereby a high- dimensional item is composed from individual elements at the encoder side and whereby a corresponding item is decomposed into conditional factors over individual elements at the decoder side, is a general method for modelling cognition involving high-dimensional items; and it suggests that the relations between the elements are best learnt in an end-to-end fashion as non-linear functions in distributed space. On the other hand, the empirical success of the networks on the tasks characterizes the underlying cognitive processes themselves: a cognitive process as complex as translating from one language to another that takes a human a few seconds to perform correctly can be accurately modelled via a learnt non-linear deterministic function of distributed vectors in high-dimensional space.
17
Bartocci, John Timothy. "Generating a synthetic dataset for kidney transplantation using generative adversarial networks and categorical logit encoding." Bowling Green State University / OhioLINK, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=bgsu1617104572023027.
Full text
18
Haidar, Ahmad. "Responsible Artificial Intelligence : Designing Frameworks for Ethical, Sustainable, and Risk-Aware Practices." Electronic Thesis or Diss., université Paris-Saclay, 2024. https://www.biblio.univ-evry.fr/theses/2024/interne/2024UPASI008.pdf.
Full textAbstract:
L'intelligence artificielle (IA) transforme rapidement le monde, redéfinissant les relations entre technologie et société. Cette thèse explore le besoin essentiel de développer, de gouverner et d'utiliser l'IA et l'IA générative (IAG) de manière responsable et durable. Elle traite des risques éthiques, des lacunes réglementaires et des défis associés aux systèmes d'IA, tout en proposant des cadres concrets pour promouvoir une Intelligence Artificielle Responsable (IAR) et une Innovation Numérique Responsable (INR).La thèse commence par une analyse approfondie de 27 déclarations éthiques mondiales sur l'IA pour identifier des principes dominants tels que la transparence, l'équité, la responsabilité et la durabilité. Bien que significatifs, ces principes manquent souvent d'outils pratiques pour leur mise en œuvre. Pour combler cette lacune, la deuxième étude de la recherche présente un cadre intégrateur pour l'IAR basé sur quatre dimensions : technique, IA pour la durabilité, juridique et gestion responsable de l'innovation.La troisième partie de la thèse porte sur l'INR à travers une étude qualitative basée sur 18 entretiens avec des gestionnaires de secteurs divers. Cinq dimensions clés sont identifiées : stratégie, défis spécifiques au numérique, indicateurs de performance organisationnels, impact sur les utilisateurs finaux et catalyseurs. Ces dimensions permettent aux entreprises d'adopter des pratiques d'innovation durable et responsable tout en surmontant les obstacles à leur mise en œuvre.La quatrième étude analyse les risques émergents liés à l'IAG, tels que la désinformation, les biais, les atteintes à la vie privée, les préoccupations environnementales et la suppression d'emplois. À partir d'un ensemble de 858 incidents, cette recherche utilise une régression logistique binaire pour examiner l'impact sociétal de ces risques. Les résultats soulignent l'urgence d'établir des cadres réglementaires renforcés, une responsabilité numérique des entreprises et une gouvernance éthique de l'IA.En conclusion, cette thèse apporte des contributions critiques aux domaines de l'INR et de l'IAR en évaluant les principes éthiques, en proposant des cadres intégratifs et en identifiant des risques émergents. Elle souligne l'importance d'aligner la gouvernance de l'IA sur les normes internationales afin de garantir que les technologies d'IA servent l'humanité de manière durable et équitable<br>Artificial Intelligence (AI) is rapidly transforming the world, redefining the relationship between technology and society. This thesis investigates the critical need for responsible and sustainable development, governance, and usage of AI and Generative AI (GAI). The study addresses the ethical risks, regulatory gaps, and challenges associated with AI systems while proposing actionable frameworks for fostering Responsible Artificial Intelligence (RAI) and Responsible Digital Innovation (RDI).The thesis begins with a comprehensive review of 27 global AI ethical declarations to identify dominant principles such as transparency, fairness, accountability, and sustainability. Despite their significance, these principles often lack the necessary tools for practical implementation. To address this gap, the second study in the research presents an integrative framework for RAI based on four dimensions: technical, AI for sustainability, legal, and responsible innovation management.The third part of the thesis focuses on RDI through a qualitative study of 18 interviews with managers from diverse sectors. Five key dimensions are identified: strategy, digital-specific challenges, organizational KPIs, end-user impact, and catalysts. These dimensions enable companies to adopt sustainable and responsible innovation practices while overcoming obstacles in implementation.The fourth study analyzes emerging risks from GAI, such as misinformation, disinformation, bias, privacy breaches, environmental concerns, and job displacement. Using a dataset of 858 incidents, this research employs binary logistic regression to examine the societal impact of these risks. The results highlight the urgent need for stronger regulatory frameworks, corporate digital responsibility, and ethical AI governance. Thus, this thesis provides critical contributions to the fields of RDI and RAI by evaluating ethical principles, proposing integrative frameworks, and identifying emerging risks. It emphasizes the importance of aligning AI governance with international standards to ensure that AI technologies serve humanity sustainably and equitably
19
Lagerkvist, Love. "Neural Novelty — How Machine Learning Does Interactive Generative Literature." Thesis, Malmö universitet, Fakulteten för kultur och samhälle (KS), 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:mau:diva-21222.
Full textAbstract:
Every day, machine learning (ML) and artificial intelligence (AI) embeds itself further into domestic and industrial technologies. Interaction de- signers have historically struggled to engage directly with the subject, facing a shortage of appropriate methods and abstractions. There is a need to find ways though which interaction design practitioners might integrate ML into their work, in order to democratize and diversify the field. This thesis proposes a mode of inquiry that considers the inter- active qualities of what machine learning does, as opposed the tech- nical specifications of what machine learning is. A shift in focus from the technicality of ML to the artifacts it creates allows the interaction designer to situate its existing skill set, affording it to engage with ma- chine learning as a design material. A Research-through-Design pro- cess explores different methodological adaptions, evaluated through user feedback and an-in depth case analysis. An elaborated design experiment, Multiverse, examines the novel, non-anthropomorphic aesthetic qualities of generative literature. It prototypes interactions with bidirectional literature and studies how these transform the reader into a cybertextual “user-reader”. The thesis ends with a discussion on the implications of machine written literature and proposes a number of future investigations into the research space unfolded through the prototype.
20
Flaherty, Drew. "Artistic approaches to machine learning." Thesis, Queensland University of Technology, 2020. https://eprints.qut.edu.au/200191/1/Drew_Flaherty_Thesis.pdf.
Full textAbstract:
This research is about how Artificial Intelligence and Machine Learning may impact creative practice. The thesis looks at various implementations and models related to the subject from different cultural and technical viewpoints. The project also provides experimental creative outcomes from my personal practice along with a qualitative study into attitudes and perspectives from other creative practitioners.
21
Townsend, Joseph Paul. "Artificial development of neural-symbolic networks." Thesis, University of Exeter, 2014. http://hdl.handle.net/10871/15162.
Full textAbstract:
Artificial neural networks (ANNs) and logic programs have both been suggested as means of modelling human cognition. While ANNs are adaptable and relatively noise resistant, the information they represent is distributed across various neurons and is therefore difficult to interpret. On the contrary, symbolic systems such as logic programs are interpretable but less adaptable. Human cognition is performed in a network of biological neurons and yet is capable of representing symbols, and therefore an ideal model would combine the strengths of the two approaches. This is the goal of Neural-Symbolic Integration [4, 16, 21, 40], in which ANNs are used to produce interpretable, adaptable representations of logic programs and other symbolic models. One neural-symbolic model of reasoning is SHRUTI [89, 95], argued to exhibit biological plausibility in that it captures some aspects of real biological processes. SHRUTI's original developers also suggest that further biological plausibility can be ascribed to the fact that SHRUTI networks can be represented by a model of genetic development [96, 120]. The aims of this thesis are to support the claims of SHRUTI's developers by producing the first such genetic representation for SHRUTI networks and to explore biological plausibility further by investigating the evolvability of the proposed SHRUTI genome. The SHRUTI genome is developed and evolved using principles from Generative and Developmental Systems and Artificial Development [13, 105], in which genomes use indirect encoding to provide a set of instructions for the gradual development of the phenotype just as DNA does for biological organisms. This thesis presents genomes that develop SHRUTI representations of logical relations and episodic facts so that they are able to correctly answer questions on the knowledge they represent. The evolvability of the SHRUTI genomes is limited in that an evolutionary search was able to discover genomes for simple relational structures that did not include conjunction, but could not discover structures that enabled conjunctive relations or episodic facts to be learned. Experiments were performed to understand the SHRUTI fitness landscape and demonstrated that this landscape is unsuitable for navigation using an evolutionary search. Complex SHRUTI structures require that necessary substructures must be discovered in unison and not individually in order to yield a positive change in objective fitness that informs the evolutionary search of their discovery. The requirement for multiple substructures to be in place before fitness can be improved is probably owed to the localist representation of concepts and relations in SHRUTI. Therefore this thesis concludes by making a case for switching to more distributed representations as a possible means of improving evolvability in the future.
22
Ladrón, de Guevara Cortés Rogelio. "Techniques For Estimating the Generative Multifactor Model of Returns in a Statistical Approach to the Arbitrage Pricing Theory. Evidence from the Mexican Stock Exchange." Doctoral thesis, Universitat de Barcelona, 2016. http://hdl.handle.net/10803/386545.
Full textAbstract:
This dissertation focuses on the estimation of the generative multifactor model of returns on equities, under a statistical approach of the Arbitrage Pricing Theory (APT), in the context of the Mexican Stock Exchange. Therefore, this research takes as frameworks two main issues: (i) the multifactor asset pricing models, specially the statistical risk factors approach, and (ii) the dimension reduction or feature extraction techniques: Principal Component Analysis, Factor Analysis, Independent Component Analysis and Non-linear Principal Component Analysis, utilized to extract the underlying systematic risk factors. The models estimated are tested using two methodologies: (i) capability of reproduction of the observed returns using the estimated generative multifactor model, and (ii) results of the econometric contrast of the APT using the extracted systematic risk factors. Finally, a comparative study among techniques is carried on based on their theoretical properties and the empirical results.
According to the above stated and as far as we concerned, this dissertation contributes to financial research by providing empirical evidence of the estimation of the generative multifactor model of returns on equities, extracting statistical underlying risk factors via classic and alternative dimension reduction or feature extraction techniques in the field of finance, in order to test the APT as an asset pricing model, in the context of an emerging financial market such as the Mexican Stock Exchange. In addition, this work presents an unprecedented theoretical and empirical comparative study among Principal Component Analysis, Factor Analysis, Independent Component Analysis and Neural Networks Principal Component Analysis, as techniques to extract systematic risk factors from a stock exchange, analyzing the level of sensitivity of the results in function of the technique carried on. In addition, this dissertation represents a mainly empirical exhaustive study where objective evidence about the Mexican stock market is provided by way of the application of four different techniques for extraction of systematic risk factors, to four datasets, in a test window that ranged from two to nine factors.
23
TOMA, ANDREA. "PHY-layer Security in Cognitive Radio Networks through Learning Deep Generative Models: an AI-based approach." Doctoral thesis, Università degli studi di Genova, 2020. http://hdl.handle.net/11567/1003576.
Full textAbstract:
Recently, Cognitive Radio (CR) has been intended as an intelligent radio endowed with cognition which can be developed by implementing Artificial Intelligence (AI) techniques. Specifically, data-driven Self-Awareness (SA) functionalities, such as detection of spectrum abnormalities, can be effectively implemented as shown by the proposed research. One important application is PHY-layer security since it is essential to establish secure wireless communications against external jamming attacks.
In this framework, signals are non-stationary and features from such kind of dynamic spectrum, with multiple high sampling rate signals, are then extracted through the Stockwell Transform (ST) with dual-resolution which has been proposed and validated in this work as part of spectrum sensing techniques.
Afterwards, analysis of the state-of-the-art about learning dynamic models from observed features describes theoretical aspects of Machine Learning (ML). In particular, following the recent advances of ML, learning deep generative models with several layers of non-linear processing has been selected as AI method for the proposed spectrum abnormality detection in CR for a brain-inspired, data-driven SA.
In the proposed approach, the features extracted from the ST representation of the wideband spectrum are organized in a high-dimensional generalized state vector and, then, a generative model is learned and employed to detect any deviation from normal situations in the analysed spectrum (abnormal signals or behaviours). Specifically, conditional GAN (C-GAN), auxiliary classifier GAN (AC-GAN), and deep VAE have been considered as deep generative models.
A dataset of a dynamic spectrum with multi-OFDM signals has been generated by using the National Instruments mm-Wave Transceiver which operates at 28 GHz (central carrier frequency) with 800 MHz frequency range. Training of the deep generative model is performed on the generalized state vector representing the mmWave spectrum with normality pattern without any malicious activity. Testing is based on new and independent data samples corresponding to abnormality pattern where the moving signal follows a different behaviour which has not been observed during training.
An abnormality indicator is measured and used for the binary classification (normality hypothesis otherwise abnormality hypothesis), while the performance of the generative models is evaluated and compared through ROC curves and accuracy metrics.
24
Ghosh, Aishik. "Simulation of the ATLAS electromagnetic calorimeter using generative adversarial networks and likelihood-free inference of the offshell Higgs boson couplings at the LHC." Thesis, université Paris-Saclay, 2020. http://www.theses.fr/2020UPASP058.
Full textAbstract:
Depuis la découverte du boson de Higgs en 2012, les expériences du LHC testent les prévisions du modèle standard avec des mesures de haute précision. Les mesures des couplages du boson de Higgs hors résonance permettront d'éliminer certaines dégénérescences qui ne peuvent pas être résolues avec les mesures sur résonance, comme la sonde de la largeur du boson de Higgs, ce qui pourrait donner des indications pour la nouvelle physique. Une partie de cette thèse se concentre sur la mesure des couplages hors résonance du boson de Higgs produit par la fusion du boson vecteur et se décomposant en quatre leptons. Ce canal de désintégration offre une occasion unique de sonder le boson de Higgs dans son régime hors résonance grâce à des sections efficaces augmentées au-delà de 2Mz (deux fois la masse du boson Z) de la région des quatre leptons. L'importante interférence quantique entre le signal et les processus de fond rend le concept d'"étiquettes de classe" mal défini, et pose un défi aux méthodes traditionnelles et aux modèles génériques de classification par apprentissage machine utilisés pour optimiser une mesure de la force du signal. Une nouvelle famille de stratégies d'inférence sans fonction de vraisemblance basées sur l'apprentissage machine, qui exploitent des informations supplémentaires pouvant être extraites du simulateur, a été adaptée à un problème de mesure de la force du signal. L'étude montre des résultats prometteurs par rapport aux techniques de base sur un ensemble de données de simulation rapide avec Delphes. Dans ce contexte, on a également introduit le réseau aspiration, un algorithme d'adverse amélioré pour la formation tout en maintenant l'invariance par rapport aux caractéristiques choisies. Les mesures de l'expérience ATLAS reposent sur de grandes quantités de données simulées précisément. Le logiciel de simulation actuel de Geant4 est trop coûteux en termes de calculs pour supporter la grande quantité de données simulées nécessaires aux analyses futures prévues. Autre partie de cette thèse se concentre sur une nouvelle approche de la simulation rapide utilisant un réseau advers génératif (GAN). La simulation de gerbe en cascade du calorimètre complexe d'ATLAS est la partie la plus lente de la chaîne de simulation utilisant Geant4. Son remplacement par un réseau de neurones qui a appris la distribution de probabilité des gerbes de particules en fonction des propriétés des particules incidentes et de la géométrie locale du détecteur augmente la vitesse de simulation de plusieurs ordres de grandeur, même sur des CPU à cœur unique, et ouvre la porte à une accélération supplémentaire sur les GPU. L'intégration dans le logiciel ATLAS permet pour la première fois de faire des comparaisons réalistes avec des simulations rapides paramétrées "à la main''. L'étude est réalisée sur une petite section du détecteur (0,20<|η|<0,25) en utilisant des photons et compare les distributions en utilisant des échantillons simulés par le modèle autonome ainsi qu'après intégration dans le logiciel ATLAS avec des échantillons Geant4 entièrement simulés. Des leçons importantes sur les mérites et les inconvénients des différentes stratégies, profitent à l'objectif ultime de simuler l'ensemble du calorimètre ATLAS avec des modèles générateurs profonds. L'étude révèle également un problème inhérent à le GAN de Wasserstein basé sur une pénalité de gradient, et propose une solution<br>Since the discovery of the Higgs boson in 2012, experiments at the LHC have been testing Standard Model predictions with high precision measurements. Measurements of the off-shell couplings of the Higgs boson will remove certain degeneracies that cannot be resolved with the current on-shell measurements, such as probing the Higgs boson width, which may lead to hints for new physics. One part of this thesis focuses on the measurement of the off-shell couplings of the Higgs boson produced by vector boson fusion and decaying to four leptons. This decay channel provides a unique opportunity to probe the Higgs in its off-shell regime due to enhanced cross-sections beyond 2Mz (twice the mass of the Z boson) region of the four lepton mass. The significant quantum interference between the signal and background processes renders the concept of `class labels' ill-defined, and poses a challenge to traditional methods and generic machine learning classification models used to optimise a signal strength measurement. A new family of machine learning based likelihood-free inference strategies, which leverage additional information that can be extracted from the simulator, were adapted to a signal strength measurement problem. The study shows promising results compared to baseline techniques on a fast simulated Delphes dataset. Also introduced in this context is the aspiration network, an improved adversarial algorithm for training while maintaining invariance with respect to chosen features. Measurements in the ATLAS experiment rely on large amounts of precise simulated data. The current Geant4 simulation software is computationally too expensive to sustain the large amount of simulated data required for planned future analyses. The other part of this thesis focuses on a new approach to fast simulation using a Generative Adversarial Network (GAN). The cascading shower simulation of the complex ATLAS calorimeter is the slowest part of the simulation chain using Geant4. Replacing it with a neural network that has learnt the probability distribution of the particle showers as a function of the incident particle properties and local detector geometry increases the simulation speed by several orders of magnitude, even on single core CPUs, and opens to door the further speed up on GPUs. The integration into the ATLAS software allows for the first time to make realistic comparisons to hand-designed fast simulation frameworks. The study is performed on a small section of the detector (0,20<|η|<0,25) using photons and compares distributions using samples simulated by the model standalone as well as after integration into the ATLAS software against fully simulated Geant4 samples. Important lessons on the merits and demerits of various strategies, benefit the ultimate goal of simulating the entire ATLAS calorimeter with a few deep generative models. The study also reveals an inherent problem with the popular gradient penalty based Wasserstein GAN, and proposes a solution
25
Karlsson, Simon, and Per Welander. "Generative Adversarial Networks for Image-to-Image Translation on Street View and MR Images." Thesis, Linköpings universitet, Institutionen för medicinsk teknik, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-148475.
Full textAbstract:
Generative Adversarial Networks (GANs) is a deep learning method that has been developed for synthesizing data. One application for which it can be used for is image-to-image translations. This could prove to be valuable when training deep neural networks for image classification tasks. Two areas where deep learning methods are used are automotive vision systems and medical imaging. Automotive vision systems are expected to handle a broad range of scenarios which demand training data with a high diversity. The scenarios in the medical field are fewer but the problem is instead that it is difficult, time consuming and expensive to collect training data. This thesis evaluates different GAN models by comparing synthetic MR images produced by the models against ground truth images. A perceptual study is also performed by an expert in the field. It is shown by the study that the implemented GAN models can synthesize visually realistic MR images. It is also shown that models producing more visually realistic synthetic images not necessarily have better results in quantitative error measurements, when compared to ground truth data. Along with the investigations on medical images, the thesis explores the possibilities of generating synthetic street view images of different resolution, light and weather conditions. Different GAN models have been compared, implemented with our own adjustments, and evaluated. The results show that it is possible to create visually realistic images for different translations and image resolutions.
26
Sveding, Jens Jakob. "Unsupervised Image-to-image translation : Taking inspiration from human perception." Thesis, Linnéuniversitetet, Institutionen för datavetenskap och medieteknik (DM), 2021. http://urn.kb.se/resolve?urn=urn:nbn:se:lnu:diva-105500.
Full textAbstract:
Generative Artificial Intelligence is a field of artificial intelligence where systems can learn underlying patterns in previously seen content and generate new content. This thesis explores a generative artificial intelligence technique used for image-toimage translations called Cycle-consistent Adversarial network (CycleGAN), which can translate images from one domain into another. The CycleGAN is a stateof-the-art technique for doing unsupervised image-to-image translations. It uses the concept of cycle-consistency to learn a mapping between image distributions, where the Mean Absolute Error function is used to compare images and thereby learn an underlying mapping between the two image distributions. In this work, we propose to use the Structural Similarity Index Measure (SSIM) as an alternative to the Mean Absolute Error function. The SSIM is a metric inspired by human perception, which measures the difference in two images by comparing the difference in, contrast, luminance, and structure. We examine if using the SSIM as the cycle-consistency loss in the CycleGAN will improve the image quality of generated images as measured by the Inception Score and Fréchet Inception Distance. The inception Score and Fréchet Inception Distance are both metrics that have been proposed as methods for evaluating the quality of images generated by generative adversarial networks (GAN). We conduct a controlled experiment to collect the quantitative metrics. Our results suggest that using the SSIM in the CycleGAN as the cycle-consistency loss will, in most cases, improve the image quality of generated images as measured Inception Score and Fréchet Inception Distance.
27
Feng, Qianli. "Modeling Action Intentionality in Humans and Machines." The Ohio State University, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=osu1616769653536292.
Full text
28
Caillon, Antoine. "Hierarchical temporal learning for multi-instrument and orchestral audio synthesis." Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS115.
Full textAbstract:
Les progrès récents en matière d'apprentissage automatique ont permis l'émergence de nouveaux types de modèles adaptés à de nombreuses tâches, ce grâce à l'optimisation d'un ensemble de paramètres visant à minimiser une fonction de coût. Parmi ces techniques, les modèles génératifs probabilistes ont permis des avancées notables dans la génération de textes, d'images et de sons. Cependant, la génération de signaux audio musicaux reste un défi. Cela vient de la complexité intrinsèque des signaux audio, une seule seconde d'audio brut comprenant des dizaines de milliers d'échantillons individuels. La modélisation des signaux musicaux est plus difficile encore, étant donné que d'importantes informations sont structurées sur différentes échelles de temps, allant du micro (timbre, transitoires, phase) au macro (genre, tempo, structure). La modélisation simultanée de toutes ces échelles implique l'utilisation de larges architectures de modèles, rendant impossible leur utilisation en temps réel en raison de la complexité de calcul. Dans cette thèse, nous proposons une approche hiérarchique de la modélisation du signal audio musical, permettant l'utilisation de modèles légers tout en offrant différents niveaux de contrôle à l'utilisateur. Notre hypothèse principale est que l'extraction de différents niveaux de représentation d'un signal audio permet d'abstraire la complexité des niveaux inférieurs pour chaque étape de modélisation. Dans un premier temps, nous proposons un modèle audio combinant Auto Encodeur Variationnel et Réseaux Antagonistes Génératifs, appliqué directement sur la forme d'onde brute et permettant une synthèse audio neuronale de haute qualité à 48 kHz, tout en étant 20 fois plus rapide que le temps réel sur CPU. Nous étudions ensuite l'utilisation d'approches autoregressives pour modéliser le comportement temporel de la représentation produite par ce modèle audio bas niveau, tout en utilisant des signaux de conditionnement supplémentaires tels que des descripteurs acoustiques ou le tempo. Enfin, nous proposons une méthode pour utiliser tous les modèles proposés directement sur des flux audio, ce qui les rend utilisables dans des applications temps réel que nous avons développées au cours de cette thèse. Nous concluons en présentant diverses collaborations créatives menées en parallèle de ce travail avec plusieurs compositeurs et musiciens, intégrant directement l'état actuel des technologies proposées au sein de pièces musicales<br>Recent advances in deep learning have offered new ways to build models addressing a wide variety of tasks through the optimization of a set of parameters based on minimizing a cost function. Amongst these techniques, probabilistic generative models have yielded impressive advances in text, image and sound generation. However, musical audio signal generation remains a challenging problem. This comes from the complexity of audio signals themselves, since a single second of raw audio spans tens of thousands of individual samples. Modeling musical signals is even more challenging as important information are structured across different time scales, from micro (e.g. timbre, transient, phase) to macro (e.g. genre, tempo, structure) information. Modeling every scale at once would require large architectures, precluding the use of resulting models in real time setups for computational complexity reasons.In this thesis, we study how a hierarchical approach to audio modeling can address the musical signal modeling task, while offering different levels of control to the user. Our main hypothesis is that extracting different representation levels of an audio signal allows to abstract the complexity of lower levels for each modeling stage. This would eventually allow the use of lightweight architectures, each modeling a single audio scale. We start by addressing raw audio modeling by proposing an audio model combining Variational Auto Encoders and Generative Adversarial Networks, yielding high-quality 48kHz neural audio synthesis, while being 20 times faster than real time on CPU. Then, we study how autoregressive models can be used to understand the temporal behavior of the representation yielded by this low-level audio model, using optional additional conditioning signals such as acoustic descriptors or tempo. Finally, we propose a method for using all the proposed models directly on audio streams, allowing their use in realtime applications that we developed during this thesis. We conclude by presenting various creative collaborations led in parallel of this work with several composers and musicians, directly integrating the current state of the proposed technologies inside musical pieces
29
Vassilo, Kyle. "Single Image Super Resolution with Infrared Imagery and Multi-Step Reinforcement Learning." University of Dayton / OhioLINK, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=dayton1606146042238906.
Full text
30
Gopinathan, Muraleekrishna. "Toward embodied navigation through vision and language." Thesis, Edith Cowan University, Research Online, Perth, Western Australia, 2025. https://ro.ecu.edu.au/theses/2894.
Full textAbstract:
Embodied AI is a challenging but exciting field in which a robot learns to interact with human-living spaces to perform various tasks. This thesis studies the embodied navigation problem in which a robotic agent navigates in a previously unseen indoor environment based on a challenging task. In particular, the Vision-and-Language Navigation (VLN) task requires a robot to navigate based on a descriptive human-language instruction. This thesis aims to improve VLN agents on four key aspects - their understanding of the environment, training via additional data, correcting navigational errors, and predicting the layout of the environment for better planning.
First, we address the planning aspect of VLN. We introduce our What Is Near (WIN) method to enhance navigation planning by predicting local neighbourhood maps using knowledge of living spaces.
Next, we study the reverse problem of VLN, where instructions are generated from trajectory demonstrations. Our Spatially-Aware Speaker (SAS) model attends to panoramic visual context and action history to decode instructions. To enhance training, a Path Mixing dataset, derived from the existing expert annotated dataset, is used and adversarial training is applied to improve instruction variety.
We observe that ambiguity in the instructions and environments leads to navigation errors and agents being lost. Our work, StratXplore, proposes the optimal navigation frontier by evaluating all available options stored in the agent’s memory based on novelty, recency, and instruction alignment.
Finally, we aim to minimise the sim-to-real gap in VLN by focusing on environment mapping in a realistic indoor simulator. The research uses the Benchbot simulator, which features a photorealistic continuous action space and realistic sensors, to map objects in indoor environments under varying conditions and sensor noises. A 2D-3D fusion pipeline is developed to evaluate state-of-the-art 3D detection models in different simulated environments. Experimental results from each study show that our methods improve upon existing work.
This thesis is an encouraging step towards realising intelligent social robots with applications in healthcare, education, and industry.
31
Mincolla, Andrea. "Space Systems of Systems Generative Design Using Concurrent MBSE: An Application of ECSS-E-TM-10-25 and the GCD Tool to Copernicus Next Generation." Thesis, KTH, Rymdteknik, 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-286332.
Full textAbstract:
The Concurrent Design Platform 4 (CDP4®) is a collaborative Model-Based Systems Engineering (MBSE) software tool conceived for architecting complex systems. Nevertheless, there are limitations concerning the manageable number of system options. The upcoming Siemens tool for generative engineering, Simcenter™ Studio, is attempting to overcome this limitation by enabling automatic synthesis and evaluation of architecture variants. The motivation for the Generative Concurrent Design (GCD) project as a collaboration between RHEA, Siemens and OHB is to develop a combined prototype of these two tools. This thesis presents a modelling of Copernicus Next Generation (CNG) space component, using generative capabilities in a typical Phase-0 study. It represents the third step of the bottom-up GCD validation process, intended to investigate how architecting differs among “Sub-system”, “Systems” and “Systems of Systems (SoS)”. Therefore, this work provides an architecting strategy which is generalizable for other SoS use-cases and contributes to assess whether extensions to ECSS-E-TM-10-25 are necessary to successfully support GCD.<br>Concurrent Design Platform 4 (CDP4®) är ett samarbetsverktyg för modellbaserad systemteknik (MBSE) som utformats för att bygga komplexa system. Dock finns det begränsningar vad gäller det antalet hanterbara systemalternativ. Det kommande Siemens-verktyget för generativ teknik, Simcenter™ Studio, försöker övervinna denna begränsning genom att möjliggöra automatisk syntes och utvärdering av arkitekturvarianter. Motivationen för Generative Concurrent Design (GCD) -projektet som ett samarbete mellan RHEA, Siemens och OHB är att utveckla en kombinerad prototyp av verktygen CDP4® och Simcenter™. Detta examensarbete presenterar en modellering av rymdkomponenten Copernicus Next Generation (CNG) med användning av generativa funktioner i en typisk fas-0-studie. Den representerar det tredje steget i GCD-valideringsprocessen nedifrån och upp, avsedd att undersöka hur arkitekturen skiljer sig åt mellan "Sub-system", "Systems" och "Systems of Systems (SoS)". Detta arbete ger därför en arkitektonisk strategi som är generaliserbar för andra SoS-användningsfall och bidrar till att bedöma om förlängningar till ECSS-E-TM-10-25 är nödvändiga för att framgångsrikt stödja GCD.
32
Rapoport, Robert S. "The iterative frame : algorithmic video editing, participant observation & the black box." Thesis, University of Oxford, 2016. https://ora.ox.ac.uk/objects/uuid:8339bcb5-79f2-44d1-b78d-7bd28aa1956e.
Full textAbstract:
Machine learning is increasingly involved in both our production and consumption of video. One symptom of this is the appearance of automated video editing applications. As this technology spreads rapidly to consumers, the need for substantive research about its social impact grows. To this end, this project maintains a focus on video editing as a microcosm of larger shifts in cultural objects co-authored by artificial intelligence. The window in which this research occurred (2010-2015) saw machine learning move increasingly into the public eye, and with it ethical concerns. What follows is, on the most abstract level, a discussion of why these ethical concerns are particularly urgent in the realm of the moving image. Algorithmic editing consists of software instructions to automate the creation of timelines of moving images. The criteria that this software uses to query a database is variable. Algorithmic authorship already exists in other media, but I will argue that the moving image is a separate case insofar as the raw material of text and music software can develop on its own. The performance of a trained actor can still not be generated by software. Thus, my focus is on the relationship between live embodied performance, and the subsequent algorithmic editing of that footage. This is a process that can employ other software like computer vision (to analyze the content of video) and predictive analytics (to guess what kind of automated film to make for a given user). How is performance altered when it has to communicate to human and non-human alike? The ritual of the iterative frame gives literal form to something that throughout human history has been a projection: the omniscient participant observer, more commonly known as the Divine. We experience black boxed software (AI's, specifically neural networks, which are intrinsically opaque) as functionally omniscient and tacitly allow it to edit more and more of life (e.g. filtering articles, playlists and even potential spouses). As long as it remains disembodied, we will continue to project the Divine on to the black box, causing cultural anxiety. In other words, predictive analytics alienate us from the source code of our cultural texts. The iterative frame then is a space in which these forces can be inscribed on the body, and hence narrated. The algorithmic editing of content is already taken for granted. The editing of moving images, in contrast, still requires a human hand. We need to understand the social power of moving image editing before it is delegated to automation. Practice Section: This project is practice-led, meaning that the portfolio of work was produced as it was being theorized. To underscore this, the portfolio comes at the end of the document. Video editors use artificial intelligence (AI) in a number of different applications, from deciding the sequencing of timelines to using facial and language detection to find actors in archives. This changes traditional production workflows on a number of levels. How can the single decision cut a between two frames of video speak to the larger epistemological shifts brought on by predictive analytics and Big Data (upon which they rely)? When predictive analytics begin modeling the world of moving images, how will our own understanding of the world change? In the practice-based section of this thesis, I explore how these shifts will change the way in which actors might approach performance. What does a gesture mean to AI and how will the editor decontextualize it? The set of a video shoot that will employ an element of AI in editing represents a move towards ritualization of production, summarized in the term the 'iterative frame'. The portfolio contains eight works that treat the set was taken as a microcosm of larger shifts in the production of culture. There is, I argue, metaphorical significance in the changing understanding of terms like 'continuity' and 'sync' on the AI-watched set. Theory Section In the theoretical section, the approach is broadly comparative. I contextualize the current dynamic by looking at previous shifts in technology that changed the relationship between production and post-production, notably the lightweight recording technology of the 1960s. This section also draws on debates in ethnographic filmmaking about the matching of film and ritual. In this body of literature, there is a focus on how participant observation can be formalized in film. Triangulating between event, participant observer and edit grammar in ethnographic filmmaking provides a useful analogy in understanding how AI as film editor might function in relation to contemporary production. Rituals occur in a frame that is dependent on a spatially/temporally separate observer. This dynamic also exists on sets bound for post-production involving AI, The convergence of film grammar and ritual grammar occurred in the 1960s under the banner of cinéma vérité in which the relationship between participant observer/ethnographer and the subject became most transparent. In Rouch and Morin's Chronicle of a Summer (1961), reflexivity became ritualized in the form of on-screen feedback sessions. The edit became transparent-the black box of cinema disappeared. Today as artificial intelligence enters the film production process this relationship begins to reverse-feedback, while it exists, becomes less transparent. The weight of the feedback ritual gets gradually shifted from presence and production to montage and post-production. Put differently, in cinéma vérité, the participant observer was most present in the frame. As participant observation gradually becomes shared with code it becomes more difficult to give it an embodied representation and thus its presence is felt more in the edit of the film. The relationship between the ritual actor and the participant observer (the algorithm) is completely mediated by the edit, a reassertion of the black box, where once it had been transparent. The crucible for looking at the relationship between algorithmic editing, participant observation and the black box is the subject in trance. In ritual trance the individual is subsumed by collective codes. Long before the advent of automated editing trance was an epistemological problem posed to film editing. In the iterative frame, for the first time, film grammar can echo ritual grammar and indeed become continuous with it. This occurs through removing the act of cutting from the causal world, and projecting this logic of post-production onto performance. Why does this occur? Ritual and specifically ritual trance is the moment when a culture gives embodied form to what it could not otherwise articulate. The trance of predictive analytics-the AI that increasingly choreographs our relationship to information-is the ineffable that finds form in the iterative frame. In the iterative frame a gesture never exists in a single instance, but in a potential state. The performers in this frame begin to understand themselves in terms of how automated indexing processes reconfigure their performance. To the extent that gestures are complicit with this mode of databasing they can be seen as votive toward the algorithmic. The practice section focuses on the poetics of this position. Chapter One focuses on cinéma vérité as a moment in which the relationship between production and post-production shifted as a function of more agile recording technology, allowing the participant observer to enter the frame. This shift becomes a lens to look at changes that AI might bring. Chapter Two treats the work of Pierre Huyghe as a 'liminal phase' in which a new relationship between production and post-production is explored. Finally, Chapter Three looks at a film in which actors perform with awareness that footage will be processed by an algorithmic edit.<br>The conclusion looks at the implications this way of relating to AI-especially commercial AI-through embodied performance could foster a more critical relationship to the proliferating black-boxed modes of production.
33
Davies, Huw. "Towards a more versatile dynamic-music for video games : approaches to compositional considerations and techniques for continuous music." Thesis, University of Oxford, 2015. http://ora.ox.ac.uk/objects/uuid:3f1e4cfa-4a36-44d8-9f4b-4c623ce6b045.
Full textAbstract:
This study contributes to practical discussions on the composition of dynamic music for video games from the composer’s perspective. Creating greater levels of immersion in players is used as a justification for the proposals of the thesis. It lays down foundational aesthetic elements in order to proceed with a logical methodology. The aim of this paper is to build upon, and further hybridise, two techniques used by composers and by video game designers to increase further the reactive agility and memorability of the music for the player. Each chapter of this paper explores a different technique for joining two (possibly disparate) types of gameplay, or gamestates, with appropriate continuous music. In each, I discuss a particular musical engine capable of implementing continuous music. Chapter One will discuss a branching-music engine, which uses a precomposed musical mosaic (or musical pixels) to create a linear score with the potential to diverge at appropriate moments accompanying onscreen action. I use the case study of the Final Fantasy battle system to show how the implementation of a branching-music engine could assist in maintaining the continuity of gameplay experience that current disjointed scores, which appear in many games, create. To aid this argument I have implemented a branching-music engine, using the graphical object oriented programming environment MaxMSP, in the style of the battle music composed by Nobuo Uematsu, the composer of the early Final Fantasy series. The reader can find this in the accompanying demonstrations patch. In Chapter Two I consider how a generative-music engine can also implement a continuous music and also address some of the limitations of the branching-music engine. Further I describe a technique for an effective generative music for video games that creates musical ‘personalities’ that can mimic a particular style of music for a limited period of time. Crucially, this engine is able to transition between any two personalities to create musical coincidence with the game. GMGEn (<b>G</b>ame <b>M</b>usic <b>G</b>eneration <b>E</b>ngine) is a program I have created in MaxMSP to act as an example of this concept. GMGEn is available in the Demonstrations_Application. Chapter Three will discuss potential limitations of the branching music engine described in Chapter One and the generative music engine described in Chapter Two, and highlights how these issues can be solved by way of a third engine, which hybridises both. As this engine has an indeterminate musical state it is termed the intermittent-music engine. I go on to discuss the implementation of this engine in two different game scenarios and how emergent structures of this music will appear. The final outcome is to formulate a new compositional approach delivering dynamic music, which accompanies the onscreen action with greater agility than currently present in the field, increasing the memorability and therefore the immersive effect of the video-game music.
34
Bordes, Florian. "Learning to sample from noise with deep generative models." Thèse, 2017. http://hdl.handle.net/1866/19370.
Full textAbstract:
L’apprentissage automatique et spécialement l’apprentissage profond se sont imposés ces
dernières années pour résoudre une large variété de tâches. Une des applications les plus
remarquables concerne la vision par ordinateur. Les systèmes de détection ou de classification ont connu des avancées majeurs grâce a l’apprentissage profond. Cependant, il reste de
nombreux obstacles à une compréhension du monde similaire aux être vivants. Ces derniers
n’ont pas besoin de labels pour classifier, pour extraire des caractéristiques du monde réel.
L’apprentissage non supervisé est un des axes de recherche qui se concentre sur la résolution
de ce problème.
Dans ce mémoire, je présente un nouveau moyen d’entrainer des réseaux de neurones de
manière non supervisée. Je présente une méthode permettant d’échantillonner de manière
itérative a partir de bruit afin de générer des données qui se rapprochent des données
d’entrainement. Cette procédure itérative s’appelle l’entrainement par infusion qui est une
nouvelle approche permettant d’apprendre l’opérateur de transition d’une chaine de Markov.
Dans le premier chapitre, j’introduis des bases concernant l’apprentissage automatique et la
théorie des probabilités. Dans le second chapitre, j’expose les modèles génératifs qui ont
inspiré ce travail. Dans le troisième et dernier chapitre, je présente comment améliorer
l’échantillonnage dans les modèles génératifs avec l’entrainement par infusion.<br>Machine learning and specifically deep learning has made significant breakthroughs in recent
years concerning different tasks. One well known application of deep learning is computer vision. Tasks such as detection or classification are nearly considered solved by the community.
However, training state-of-the-art models for such tasks requires to have labels associated
to the data we want to classify. A more general goal is, similarly to animal brains, to be
able to design algorithms that can extract meaningful features from data that aren’t labeled.
Unsupervised learning is one of the axes that try to solve this problem.
In this thesis, I present a new way to train a neural network as a generative model capable of
generating quality samples (a task akin to imagining). I explain how by starting from noise,
it is possible to get samples which are close to the training data. This iterative procedure
is called Infusion training and is a novel approach to learning the transition operator of a
generative Markov chain.
In the first chapter, I present some background about machine learning and probabilistic
models. The second chapter presents generative models that inspired this work. The third
and last chapter presents and investigates our novel approach to learn a generative model
with Infusion training.
35
Hamdi, Abdullah. "Cascading Generative Adversarial Networks for Targeted." Thesis, 2018. http://hdl.handle.net/10754/627557.
Full textAbstract:
Abundance of labelled data played a crucial role in the recent developments in computer vision, but that faces problems like scalability and transferability to the wild. One alternative approach is to utilize the data without labels, i.e. unsupervised learning, in learning valuable information and put it in use to tackle vision problems. Generative Adversarial Networks (GANs) have gained momentum for their ability to model image distributions in unsupervised manner. They learn to emulate the training set and that enables sampling from that domain and using the knowledge learned for useful applications. Several methods proposed enhancing GANs, including regularizing the loss with some feature matching. We seek to push GANs beyond the data in the training and try to explore unseen territory in the image manifold. We first propose a new regularizer for GAN based on K-Nearest Neighbor (K-NN) selective feature matching to a target set Y in high-level feature space, during the adversarial training of GAN on the base set X, and we call this novel model K-GAN. We show that minimizing the added term follows from cross-entropy minimization between the distributions of GAN and set Y. Then, we introduce a cascaded framework for GANs that try to address the task of imagining a new distribution that combines the base set X and target set Y by cascading sampling GANs with translation GANs, and we dub the cascade of such GANs as the Imaginative Adversarial Network (IAN). Several cascades are trained on a collected dataset Zoo-Faces and generated innovative samples are shown, including from K-GAN cascade. We conduct an objective and subjective evaluation for different IAN setups in the addressed task of generating innovative samples and we show the effect of regularizing GAN on different scores. We conclude with some useful applications for these IANs, like multi-domain manifold traversing.
36
(7242737), Pradeep Periasamy. "Generative Adversarial Networks for Lupus Diagnostics." Thesis, 2019.
Find full textAbstract:
The recent boom of Machine Learning Network Architectures like Generative
Adversarial Networks (GAN), Deep Convolution Generative Adversarial Networks
(DCGAN), Self Attention Generative Adversarial Networks (SAGAN), Context
Conditional Generative Adversarial Networks (CCGAN) and the development of
high-performance computing for big data analysis has the potential to be highly
beneficial in many domains and fittingly in the early detection of chronic diseases.
The clinical heterogeneity of one such chronic auto-immune disease like Systemic
Lupus Erythematosus (SLE), also known as Lupus, makes it difficult for medical
diagnostics. One major concern is a limited dataset that is available for diagnostics.
In this research, we demonstrate the application of Generative Adversarial Networks
for data augmentation and improving the error rates of Convolution Neural
Networks (CNN). Limited Lupus dataset of 30 typical ’butterfly rash’ images is used
as a model to decrease the error rates of a widely accepted CNN architecture like
Le-Net. For the Lupus dataset, it can be seen that there is a 73.22% decrease in the
error rates of Le-Net. Therefore such an approach can be extended to most recent
Neural Network classifiers like ResNet. Additionally, a human perceptual study
reveals that the artificial images generated from CCGAN are preferred to closely
resemble real Lupus images over the artificial images generated from SAGAN and
DCGAN by 45 Amazon MTurk participants. These participants are identified as
’healthcare professionals’ in the Amazon MTurk platform. This research aims to
help reduce the time in detection and treatment of Lupus which usually takes 6 to 9
months from its onset.
37
Kastner, Kyle. "Structured prediction and generative modeling using neural networks." Thèse, 2016. http://hdl.handle.net/1866/18760.
Full textAbstract:
Cette thèse traite de l'usage des Réseaux de Neurones pour modélisation de données séquentielles. La façon dont l'information a été ordonnée et structurée est cruciale pour la plupart des données. Les mots qui composent ce paragraphe en constituent un exemple. D'autres données de ce type incluent les données audio, visuelles et génomiques. La Prédiction Structurée est l'un des domaines traitant de la modélisation de ces données. Nous allons aussi présenter la Modélisation Générative, qui consiste à générer des points similaires aux données sur lesquelles le modèle a été entraîné.
Dans le chapitre 1, nous utiliserons des données clients afin d'expliquer les concepts et les outils de l'Apprentissage Automatique, incluant les algorithmes standards d'apprentissage ainsi que les choix de fonction de coût et de procédure d'optimisation. Nous donnerons ensuite les composantes fondamentales d'un Réseau de Neurones. Enfin, nous introduirons des concepts plus complexes tels que le partage de paramètres, les Réseaux Convolutionnels et les Réseaux Récurrents. Le reste du document, nous décrirons de plusieurs types de Réseaux de Neurones qui seront à la fois utiles pour la prédiction et la génération et leur application à des jeux de données audio, d'écriture manuelle et d'images.
Le chapitre 2 présentera le Réseau Neuronal Récurrent Variationnel (VRNN pour variational recurrent neural network). Le VRNN a été développé dans le but de générer des échantillons semblables aux exemples de la base d'apprentissage. Nous présenterons des modèles entraînées de manière non-supervisée afin de générer du texte manuscrites, des effets sonores et de la parole. Non seulement ces modèles prouvent leur capacité à apprendre les caractéristiques de chaque type de données mais établissent aussi un standard en terme de performance.
Dans le chapitre 3 sera présenté ReNet, un modèle récemment développé. ReNet utilise les sorties structurées d'un Réseau Neuronal Récurrent pour classifier des objets. Ce modèle atteint des performances compétitives sur plusieurs tâches de reconnaissance d'images, tout en utilisant une architecture conçue dès le départ pour de la Prédiction Structurée. Dans ce cas-ci, les résultats du modèle sont utilisés simplement pour de la classification mais des travaux suivants (non inclus ici) ont utilisé ce modèle pour de la Prédiction Structurée.
Enfin, au Chapitre 4 nous présentons les résultats récents non-publiés en génération acoustique. Dans un premier temps, nous fournissons les concepts musicaux et représentations numériques fondamentaux à la compréhension de notre approche et introduisons ensuite une base de référence et de nouveaux résultats de recherche avec notre modèle, RNN-MADE. Ensuite, nous introduirons le concept de synthèse vocale brute et discuterons de notre recherche en génération. Dans notre dernier Chapitre, nous présenterons enfin un résumé des résultats et proposerons de nouvelles pistes de recherche.<br>In this thesis we utilize neural networks to effectively model data with sequential structure.
There are many forms of data for which both the order and the structure of the information is incredibly important. The words in this paragraph are one example of this type of data. Other examples include audio, images, and genomes. The work to effectively model this type of ordered data falls within the field of structured prediction. We also present generative models, which attempt to generate data that appears similar to the data which the model was trained on.
In Chapter 1, we provide an introduction to data and machine learning. First, we motivate the need for machine learning by describing an expert system built on a customer database. This leads to a discussion of common algorithms, losses, and optimization choices in machine learning. We then progress to describe the basic building blocks of neural networks. Finally, we add complexity to the models, discussing parameter sharing and convolutional and recurrent layers. In the remainder of the document, we discuss several types of neural networks which find common use in both prediction and generative modeling and present examples of their use with audio, handwriting, and images datasets.
In Chapter 2, we introduce a variational recurrent neural network (VRNN). Our VRNN is developed with to generate new sequential samples that resemble the dataset that is was trained on. We present models that learned in an unsupervised manner how to generate handwriting, sound effects, and human speech setting benchmarks in performance.
Chapter 3 shows a recently developed model called ReNet. In ReNet, intermediate structured outputs from recurrent neural networks are used for object classification. This model shows competitive performance on a number of image recognition tasks, while using an architecture designed to handle structured prediction. In this case, the final model output is only used for simple classification, but follow-up work has expanded to full structured prediction.
Lastly, in Chapter 4 we present recent unpublished experiments in sequential audio generation. First we provide background in musical concepts and digital representation which are fundamental to understanding our approach and then introduce a baseline and new research results using our model, RNN-MADE. Next we introduce the concept of raw speech synthesis and discuss our investigation into generation. In our final chapter, we present a brief summary of results and postulate future research directions.
38
Goyal, Anirudh. "Improved training of generative models." Thèse, 2018. http://hdl.handle.net/1866/22073.
Full text
39
"Zero Shot Learning for Visual Object Recognition with Generative Models." Master's thesis, 2020. http://hdl.handle.net/2286/R.I.57065.
Full textAbstract:
abstract: Visual object recognition has achieved great success with advancements in deep learning technologies. Notably, the existing recognition models have gained human-level performance on many of the recognition tasks. However, these models are data hungry, and their performance is constrained by the amount of training data. Inspired by the human ability to recognize object categories based on textual descriptions of objects and previous visual knowledge, the research community has extensively pursued the area of zero-shot learning. In this area of research, machine vision models are trained to recognize object categories that are not observed during the training process. Zero-shot learning models leverage textual information to transfer visual knowledge from seen object categories in order to recognize unseen object categories.
Generative models have recently gained popularity as they synthesize unseen visual features and convert zero-shot learning into a classical supervised learning problem. These generative models are trained using seen classes and are expected to implicitly transfer the knowledge from seen to unseen classes. However, their performance is stymied by overfitting towards seen classes, which leads to substandard performance in generalized zero-shot learning. To address this concern, this dissertation proposes a novel generative model that leverages the semantic relationship between seen and unseen categories and explicitly performs knowledge transfer from seen categories to unseen categories. Experiments were conducted on several benchmark datasets to demonstrate the efficacy of the proposed model for both zero-shot learning and generalized zero-shot learning. The dissertation also provides a unique Student-Teacher based generative model for zero-shot learning and concludes with future research directions in this area.<br>Dissertation/Thesis<br>Masters Thesis Computer Science 2020
40
"A Study on Generative Adversarial Networks Exacerbating Social Data Bias." Master's thesis, 2020. http://hdl.handle.net/2286/R.I.57433.
Full textAbstract:
abstract: Generative Adversarial Networks are designed, in theory, to replicate the distribution of the data they are trained on. With real-world limitations, such as finite network capacity and training set size, they inevitably suffer a yet unavoidable technical failure: mode collapse. GAN-generated data is not nearly as diverse as the real-world data the network is trained on; this work shows that this effect is especially drastic when the training data is highly non-uniform. Specifically, GANs learn to exacerbate the social biases which exist in the training set along sensitive axes such as gender and race. In an age where many datasets are curated from web and social media data (which are almost never balanced), this has dangerous implications for downstream tasks using GAN-generated synthetic data, such as data augmentation for classification. This thesis presents an empirical demonstration of this phenomenon and illustrates its real-world ramifications. It starts by showing that when asked to sample images from an illustrative dataset of engineering faculty headshots from 47 U.S. universities, unfortunately skewed toward white males, a DCGAN’s generator “imagines” faces with light skin colors and masculine features. In addition, this work verifies that the generated distribution diverges more from the real-world distribution when the training data is non-uniform than when it is uniform. This work also shows that a conditional variant of GAN is not immune to exacerbating sensitive social biases. Finally, this work contributes a preliminary case study on Snapchat’s explosively popular GAN-enabled “My Twin” selfie lens, which consistently lightens the skin tone for women of color in an attempt to make faces more feminine. The results and discussion of the study are meant to caution machine learning practitioners who may unsuspectingly increase the biases in their applications.<br>Dissertation/Thesis<br>Masters Thesis Computer Science 2020
41
Alves, Diogo Rafael Cordeiro. "Modeling of Synthetic Players As An Instrument For Testing Generative Content." Master's thesis, 2021. http://hdl.handle.net/10316/96018.
Full textAbstract:
Dissertação de Mestrado em Engenharia Informática apresentada à Faculdade de Ciências e Tecnologia<br>Há uma necessidade de se encontrar novos métodos confiáveis para testar cenários de jogo gerados procedimentalmente. Um método que mostra potencial é teste automatizado, que consiste em usar jogador sintéticos para testar novos cenários de jogo. Deste modo, para testar esta suposição, este trabalho modela dois tipos de jogadores sintéticos que são colocados num cenário semelhante ao Bomberman. Neste caso de estudo, os jogadores têm de colocar bombas em lugares estratégicos para eliminar os adversários. Um jogador sintético é desenvolvido através de uma abordagem baseada em planeamento, em que o agente procura por uma sequência de ações que leva desde o estado atual do jogo até ao estado desejado. A segunda abordagem é aprendizagem computacional, mais precisamente combinar aprendizagem por imitação com aprendizagem por reforço, para que primeiro, o jogador sintético aprenda observando demonstrações humanas, e depois melhore a sua performance através das recompensas do ambiente. Resultados mostram que o jogador sintético de planeamento consegue jogar e ganhar o jogo consistentemente contra oponentes desenvolvidos pela abordagem de aprendizagem computacional, e consegue generalizar bem para novos cenários. Também obteve resultados positivos num inquérito acerca da sua credibilidade. Estes atributos fazem do jogador sintético de planeamento uma ferramente viável para testar cenários de jogo gerados procedimentalmente. Contudo, o agente de planeamento tem dificuldades quando joga contra um jogador humano, fornecendo apenas um desafio de dificuldade fácil a moderada para o humano. A abordagem de aprendizagem computacional produziu resultados modestos, não sendo capaz de vencer o jogo consistentemente. O seu desempenho pouco satisfatório também magoa a sua credibilidade. Não obstante esse facto, um modelo treinado aparenta ter aprendido as regras básicas do jogo e conseguiu sobreviver períodos de tempo suficientes para explorar o espaço de jogo.<br>There is a need to discover new, reliable techniques to test procedurally generated game scenarios. One method that has potential is automated testing, which consists in using synthetic players to play and test the newly generated scenarios. Therefore, to test this assumption, this work models two types of synthetic players that are put in a Bomberman-like scenario. In this case study, the players must place bombs in strategic places in order to eliminate the opponents. One synthetic player is developed via a planning-based approach, in which the agent searches for a sequence of actions that leads from the current state of the game to the desired game state. The second approach is machine learning, more precisely, combining Imitation Learning with Reinforcement Learning in order for the synthetic player to first learn by observing human demonstrations, and then improve its performance by maximizing its policy via the environment rewards. Results show that the planning synthetic player manages to play and win the game consistently against opponents developed by the machine learning approach, and can generalize well to new unseen scenarios. It also obtained positive scores in a survey regarding its believability. All of these attributes makes the planning synthetic player a viable tool to test procedurally generated game scenarios. However, the planning agent struggles when facing a human player, providing an easy-to-moderate challenge for the human. The machine learning approach produced modest results, not being able to win the game consistently. Its modest performance harms its believability as well. Nonetheless, a particular trained model could be perceived as if grasping how to play the game and managed to survive long-enough periods to explore the game space.
42
Burjorjee, Keki M. "Generative fixation : a unified explanation for the adaptive capacity of simple recombinative genetic algorithms /." 2009.
Find full text
43
Mehri, Soroush. "Sequential modeling, generative recurrent neural networks, and their applications to audio." Thèse, 2016. http://hdl.handle.net/1866/18762.
Full text
44
Belghazi, Mohamed. "Estimation neuronale de l'information mutuelle." Thesis, 2020. http://hdl.handle.net/1866/25093.
Full textAbstract:
Nous argumentons que l'estimation de l'information mutuelle entre des ensembles de variables aléatoires continues de hautes dimensionnalités peut être réalisée par descente de gradient sur des réseaux de neurones.
Nous présentons un estimateur neuronal de l'information mutuelle (MINE) dont la complexité croît linéairement avec la dimensionnalité des variables et la taille de l'échantillon, entrainable par retro-propagation, et fortement consistant au sens statistique. Nous présentons aussi une poignée d'application ou MINE peut être utilisé pour minimiser ou maximiser l'information mutuelle. Nous appliquons MINE pour améliorer les modèles génératifs adversariaux. Nous utilisons aussi MINE pour implémenter la méthode du goulot d'étranglement de l'information dans un cadre de classification supervisé. Nos résultats montrent un gain substantiel en flexibilité et performance.<br>We argue that the estimation of mutual information between high dimensional continuous random variables can be achieved by gradient descent over neural networks. We present a Mutual Information Neural Estimator (MINE) that is linearly scalable in dimensionality as well as in sample size, trainable through back-prop, and strongly consistent. We present a handful of applications on which MINE can be used to minimize or maximize mutual information. We apply MINE to improve adversarially trained generative models. We also use MINE to implement the Information Bottleneck, applying it to supervised classification; our results demonstrate substantial improvement in flexibility and performance in the settings.
45
Capobianco, Samuele. "Deep Learning Methods for Document Image Understanding." Doctoral thesis, 2020. http://hdl.handle.net/2158/1182536.
Full textAbstract:
Document image understanding involves several tasks including, among others, the layout analysis of historical handwritten and the symbol recognition in graphical documents. The understanding of document images implies two
processes, the analysis, and the recognition, which are complex tasks. Moreover, each application domain has a specific information structure which increases the complexity of the understanding process.
In the last years, many machine learning approaches have been presented to address document image understanding. In this research, we present a series of deep learning methods to address different application domains: historical handwritten and graphical documents understanding. We show the difficulties encountered when applying these techniques and the proposed solutions for each application domain. We cope with the problem of working with supervised deep networks that require to have a large dataset for training. We address the over-fitting related to the scarcity of labeled data showing several solutions to prevent this issue in these application domains.
First, we show our contributions to historical handwritten layout analysis. We propose a toolkit to generate structured synthetic documents emulating the actual document production process. Synthetic documents can be used to train systems to perform layout analysis. Then, we study the use of deep networks for counting the number of records in each page of a historical handwritten document. Furthermore, we present a novel approach for the extraction of text lines in handwritten documents using another deep network to label document image patches as text lines or separators. Related to the page segmentation, we propose a fully convolutional network trained by a domain-specific loss for classifying pixels to segment semantic regions on handwritten pages. Second, we propose a novel interactive annotation system to help users to label symbols at the pixel level for the graphical symbol understanding problem. Using the proposed interactive system we can improve the annotation results and reduce the time-consuming process of labeling data. Using this system, we built a novel floor plan image dataset for object detection. We show preliminary results by using state-of-the-art deep networks to detect symbols on this dataset. In the end, we provide an extensive discussion for each task addressed showing the obtained results and proposing future works.
46
Mastropietro, Olivier. "Deep Learning for Video Modelling." Thèse, 2017. http://hdl.handle.net/1866/20192.
Full text
47
Almahairi, Amjad. "Advances in deep learning with limited supervision and computational resources." Thèse, 2018. http://hdl.handle.net/1866/23434.
Full textAbstract:
Les réseaux de neurones profonds sont la pierre angulaire des systèmes à la fine pointe de la technologie pour une vaste gamme de tâches, comme la reconnaissance d'objets, la modélisation du langage et la traduction automatique.
Mis à part le progrès important établi dans les architectures et les procédures de formation des réseaux de neurones profonds, deux facteurs ont été la clé du succès remarquable de l'apprentissage profond : la disponibilité de grandes quantités de données étiquetées et la puissance de calcul massive.
Cette thèse par articles apporte plusieurs contributions à l'avancement de l'apprentissage profond,
en particulier dans les problèmes avec très peu ou pas de données étiquetées, ou avec des ressources informatiques limitées.
Le premier article aborde la question de la rareté des données dans les systèmes de recommandation, en apprenant les représentations distribuées des produits à partir des commentaires d'évaluation de produits en langage naturel. Plus précisément, nous proposons un cadre d'apprentissage multitâches dans lequel nous utilisons des méthodes basées sur les réseaux de neurones pour apprendre les représentations de produits à partir de textes de critiques de produits et de données d'évaluation. Nous démontrons que la méthode proposée peut améliorer la généralisation dans les systèmes de recommandation et atteindre une performance de pointe sur l'ensemble de données Amazon Reviews.
Le deuxième article s'attaque aux défis computationnels qui existent dans l'entraînement des réseaux de neurones profonds à grande échelle. Nous proposons une nouvelle architecture de réseaux de neurones conditionnels permettant d'attribuer la capacité du réseau de façon adaptative, et donc des calculs, dans les différentes régions des entrées. Nous démontrons l'efficacité de notre modèle sur les tâches de reconnaissance visuelle où les objets d'intérêt sont localisés à la couche d'entrée, tout en maintenant une surcharge de calcul beaucoup plus faible que les architectures standards des réseaux de neurones.
Le troisième article contribue au domaine de l'apprentissage non supervisé, avec l'aide du paradigme des réseaux antagoniste génératifs. Nous introduisons un cadre fléxible pour l'entraînement des réseaux antagonistes génératifs, qui non seulement assure que le générateur estime la véritable distribution des données, mais permet également au discriminateur de conserver l'information sur la densité des données à l'optimum global. Nous validons notre cadre empiriquement en montrant que le discriminateur est capable de récupérer l'énergie de la distribution des données et d'obtenir une qualité d'échantillons à la fine pointe de la technologie.
Enfin, dans le quatrième article, nous nous attaquons au problème de l'apprentissage non supervisé à travers différents domaines. Nous proposons un modèle qui permet d'apprendre des transformations plusieurs à plusieurs à travers deux domaines, et ce, à partir des données non appariées. Nous validons notre approche sur plusieurs ensembles de données se rapportant à l'imagerie, et nous montrons que notre méthode peut être appliquée efficacement dans des situations d'apprentissage semi-supervisé.<br>Deep neural networks are the cornerstone of state-of-the-art systems for a wide range of tasks, including object recognition, language modelling and machine translation. In the last decade, research in the field of deep learning has led to numerous key advances in designing novel architectures and training algorithms for neural networks. However, most success stories in deep learning heavily relied on two main factors: the availability of large amounts of labelled data and massive computational resources.
This thesis by articles makes several contributions to advancing deep learning,
specifically in problems with limited or no labelled data, or with constrained computational resources.
The first article addresses sparsity of labelled data that emerges in the application field of recommender systems. We propose a multi-task learning framework that leverages natural language reviews in improving recommendation.
Specifically, we apply neural-network-based methods for learning representations of products from review
text, while learning from rating data. We demonstrate that the proposed method can achieve state-of-the-art performance
on the Amazon Reviews dataset.
The second article tackles computational challenges in training large-scale
deep neural networks. We propose a conditional computation network
architecture which can adaptively assign its capacity, and hence computations,
across different regions of the input. We demonstrate the effectiveness of
our model on visual recognition tasks where objects are spatially localized within the input, while maintaining much lower computational overhead than standard network architectures.
The third article contributes to the domain of unsupervised learning with the generative adversarial networks paradigm. We introduce a flexible adversarial training framework, in which not only the generator converges to the true data distribution, but also the discriminator recovers the relative density of the data at the optimum. We validate our framework empirically by showing that the discriminator is able to accurately estimate the true energy of data while obtaining state-of-the-art quality of samples.
Finally, in the fourth article, we address the problem of unsupervised
domain translation. We propose a model which can learn
flexible, many-to-many mappings across domains from unpaired data. We validate
our approach on several image datasets, and we show that it can be effectively applied in semi-supervised learning settings.
48
Rodrigues, Diogo Manuel de Castro. "Integrating Vision and Language for Automatic Face Descriptions." Master's thesis, 2018. http://hdl.handle.net/10316/86752.
Full textAbstract:
Dissertação de Mestrado Integrado em Engenharia Electrotécnica e de Computadores apresentada à Faculdade de Ciências e Tecnologia<br>Nesta dissertação, para criar um exemplo único de um sistema de face para texto e texto para face foi integrado visão por computador e processamento de linguagem natural. O propósito é fornecer uma solução que permita ajudar os seres humanos a realizar funções com maior qualidade e de forma mais rápida. Assim sendo pretende-se criar um sistema que possa ser usado, por exemplo, para descrever rostos para pessoas com deficiência visual ou para gerar rostos a partir de descrições para investigações criminais. No entanto trata-se apenas de uma versão preliminar, na medida em que o curto tempo disponível para a sua realização não permitiu alcançar a ambiciosa proposta. De forma a atingir este objectivo, foi criado um sistema com a capacidade de descrever textualmente imagens faciais e por outro lado, gerar automaticamente imagens faciais a partir de descrições textuais. O sistema é dividido em duas partes, a primeira tem como função prever atributos das imagens faciais através de uma rede neuronal convolucional. Estes são utilizados como base para o modelo de geração de linguagem natural, gerando descrições textuais numa metodologia baseada em regras. A segunda parte, usa uma técnica simples de extração de palavras chave para analisar o texto e identificar os atributos nessa descrição. Seguidamente, o sistema usa uma rede generativa adversarial para gerar uma imagem facial com o conjunto das características desejadas. Os atributos são usados como base no nosso método, uma vez que representam um identificador dominante que transmite características sobre um rosto com eficácia.Os resultados demonstraram, mais uma vez, que os métodos CNN e GAN são atualmente as melhores opções para, tarefas de reconhecimento e geração de imagens, respectivamente. Esta conclusão destá assente nos resultados convincentes. Por outro lado, os métodos de processamento de linguagem natural apesar de terem funcionado bem, de acordo com os objectivos, os seus resultados são menos notáveis, especialmente o modelo de geração de linguagem natural. Este trabalho propõe uma solução fiável e funcional para resolver este sistema complexo, no entanto é uma área que merece uma extensa investigação e desenvolvimento.<br>In this dissertation, computer vision and Natural Language Processing (NLP) are integrated to create a unique example of a face-to-text and text-to-face system. Its intention is to provide a solution that can help humans to perform their jobs with better quality and with a quick response. The aim is to create a system that can be used, for example, to describe faces for visually impaired people or to generate faces from descriptions for criminal investigations. However, this is a preliminary version as it is an ambitious goal to be achieved during the time available for its realization.To accomplish this motivation, a system was created with the capability of describing, textually, facial images, along with the ability to automatically generate face images from text descriptions. The system is divided into two sub-systems. The first part predicts attributes from the face images through a Convolutional Neural Network (CNN) method that are used, further, as a base to the Natural Language Generation (NLG) model. The descriptions are generated on a rule-based methodology. The second part of the system uses a simple keyword extraction technique to analyze the text and identify the attributes on that description. After that, it uses a conditional Generative Adversarial Network (GAN) to generate a facial image with a specific set of desired attributes. The reason why attributes are used as a base on the method is because they are a dominant identifier that can efficiently transmit characteristic about a face. The results demonstrate, once again, that either CNN and GAN methods are presently the best options for recognition and generation tasks, respectively. This conclusion is due to their convincing results. On the other hand, the NLP methods worked well for their purposes. However, its results are less remarkable, especially the NLG model. This work proposes a reliable and functional solution for solving this complex system. Nevertheless, this area needs an extensive investigation and development.
49
Dutil, Francis. "Prédiction et génération de données structurées à l'aide de réseaux de neurones et de décisions discrètes." Thèse, 2018. http://hdl.handle.net/1866/22124.
Full text
50
Kumar, Rithesh. "Improved training of energy-based models." Thèse, 2019. http://hdl.handle.net/1866/22528.
Full text