Academic literature on the topic 'IGNGF'

In a real world application, we seldom get all images at one time. Considering this case, if a company hired an employee, all his images information needs to be recorded into the system; if we rerun the face recognition algorithm, it will be time consuming. To address this problem, In this paper, firstly, we proposed a novel subspace incremental method called incremental graph regularized nonnegative matrix factorization (IGNMF) algorithm which imposes manifold into incremental nonnegative matrix factorization algorithm (INMF); thus, our new algorithm is able to preserve the geometric structure in the data under incremental study framework; secondly, considering we always get many face images belonging to one person or many different people as a batch, we improved our IGNMF algorithms to Batch-IGNMF algorithms (B-IGNMF), which implements incremental study in batches. Experiments show that (1) the recognition rate of our IGNMF and B-IGNMF algorithms is close to GNMF algorithm while it runs faster than GNMF. (2) The running times of our IGNMF and B-IGNMF algorithms are close to INMF while the recognition rate outperforms INMF. (3) Comparing with other popular NMF-based face recognition incremental algorithms, our IGNMF and B-IGNMF also outperform then both the recognition rate and the running time.

We investigated the influence of three different high-pass (HP) and low-pass (LP) filtering conditions and a Gaussian (GNMF) and inverse-Gaussian (IGNMF) non-negative matrix factorization algorithm on the extraction of muscle synergies from myoelectric signals during human walking and running. To evaluate the effects of signal recording and processing on the outcomes, we analyzed the intraday and interday computation reliability. Results show that the IGNMF achieved a significantly higher reconstruction quality and on average needs one less synergy to sufficiently reconstruct the original signals compared to the GNMF. For both factorizations, the HP with a cut-off frequency of 250[Formula: see text]Hz significantly reduces the number of synergies. We identified the filter configuration of fourth order, HP 50[Formula: see text]Hz and LP 20[Formula: see text]Hz as the most suitable to minimize the combination of fundamental synergies, providing a higher reliability across all filtering conditions even if HP 250[Formula: see text]Hz is excluded. Defining a fundamental synergy as a single-peaked activation pattern, for walking and running we identified five and six fundamental synergies, respectively using both algorithms. The variability in combined synergies produced by different filtering conditions and factorization methods on the same data set suggests caution when attributing a neurophysiological nature to the combined synergies.

In this study, three months of records (January–March 2010) that were acquired by a geodetic Global Navigation Satellite Systems (GNSS) station from the permanent network of RGP (Réseau GNSS Permanent), which was deployed by the French Geographic Institute (IGNF), located in Socoa, in the south of the Bay of Biscay, were used to determine the tide components and identify the signature of storms on the signal to noise ratio (SNR) during winter 2010. The Xynthia storm hit the French Atlantic coast on the 28th of February 2010, causing large floods and damages from the Gironde to the Loire estuaries. Blind separation of the tide components and of the storm signature was achieved while using both a singular spectrum analysis (SSA) and a continuous wavelet transform (CWT). A correlation of 0.98/0.97 and root mean square error (RMSE) of 0.21/0.28 m between the tide gauge records of Socoa and our estimates of the sea surface height (SSH) using the SSA and the CWT, respectively, were found. Correlations of 0.76 and 0.7 were also obtained between one of the modes from the SSA and atmospheric pressure from a meteorological station and a mode of the SSA. Particularly, a correlation reaches to 0.76 when using both the tide residual that is associated to surges and atmospheric pressure variation.

Background: While autologous stem cell transplantation (ASCT) can be curative for patients (pts) with relapsed or refractory diffuse large B-cell lymphoma (DLBCL), relapse remains common. With the emergence of novel effective therapies, it is even more important to identify pts at high risk of treatment failure who may not benefit from ASCT, and pts with impending post-ASCT relapse who may be candidates for pre-emptive interventions. We assembled cohorts of DLBCL pts who underwent ASCT and had apheresis stem cell (ASC) samples or serially collected post-ASCT peripheral blood mononuclear cell (PBMC) and plasma samples. We hypothesized that circulating tumor DNA (ctDNA) identified using immunoglobulin-based next generation sequencing (IgNGS) in ASC or PB samples could predict relapse. Methods: Samples from 3 cohorts were analyzed. Pts in cohort 1 (C1) underwent ASCT at Dana-Farber Cancer Institute (DFCI) from 2003-2013 (Herrera, ASH 2015). Archival tumor tissue and ASC samples were retrospectively collected for analysis. Pts in cohort 2 (C2) were prospectively enrolled on a banking protocol at DFCI and underwent ASCT from 2014-2016. Pts in cohort 3 (C3) underwent ASCT from 2015-2016 and participated in a multicenter phase II trial of post-ASCT pembrolizumab maintenance (PM) (Frigault, Blood Adv 2020). Pts in C2/C3 had tumor tissue and serially collected post-ASCT PBMC and plasma samples as mandated by protocol, and a subset had available pre-ASCT PB or ASC samples. Because PM did not demonstrate a clear benefit in the trial, all cohorts were analyzed together. IgNGS (Adaptive Biotechnologies; Seattle, WA) was performed, as previously described (Armand, BJH 2013). In all cases, ctDNA testing was not performed in real-time or used to drive clinical decisions. Results: 152 pts were enrolled. Among 141 pts with sufficient DNA for testing, a clonotype was identified in 112 (78%) with a higher detection rate in more recent cohorts - C2 (93%) and C3 (90%) vs C1 (67%). Among 97 pts with an available ASC sample, 23 (24%) were ctDNA-positive (pos). With a median follow-up among survivors of 69 months (m) (range 13-185), the 5-year (y) progression-free survival (PFS) for ASC ctDNA-pos and ASC ctDNA-negative (neg) pts were 13% (95% CI 3-30%) and 52% (95% CI 40-63%), respectively (HR 2.8, p<0.001), while the 5y cumulative incidences of relapse were 83% (95% CI 66-99%) and 39% (95% CI 27-50%), respectively (HR 3.1, p<0.001). The sensitivity and specificity of ASC ctDNA for progression or death were 36% and 95%, respectively. ASC ctDNA (HR 2.5, p=0.002) was the only significant predictor of PFS in a multivariable model that included pre-ASCT positron emission tomography (PET), lines of therapy, age, and primary refractory status. Inferior overall survival was observed for ASC ctDNA-pos pts (HR 2.1 p=0.037). In an exploratory analysis, we examined 14 pts with an available pre-ASCT plasma sample. 2/14 were ctDNA-pos (14%) and both pts relapsed (HR for PFS 9.4, p=0.03). Among 13 pts with both pre-ASCT PB and ASC samples (drawn a median of 19 days apart [range 11-47]), results were concordant in 12/13 pts (92%). 56 pts had a median of 3 (range 1-8) post-ASCT plasma samples available for analysis. Within this cohort, 25 pts relapsed and 2 pts died in remission. 21 pts (38%) had detectable ctDNA in a median of 2 post-ASCT samples (range 1-5); among them, 18 (86%) relapsed with a median lead time from first ctDNA detection to relapse of 52 days (range 0-518). Among the 3 ctDNA-pos pts who did not relapse, 2 had detectable ctDNA at a single time point and subsequently became ctDNA-neg, and 1 developed acute myeloid leukemia and underwent allogeneic transplantation. Among 20 pts who relapsed and had ≥1 plasma sample available within 100 days of relapse, 18 (90%) had detectable ctDNA. PBMC testing had inferior performance characteristics (Table). Conclusions: Identification of ctDNA using IgNGS within an ASC sample is a powerful predictor of post-ASCT relapse and provides (at least in this cohort) a better way to predict relapse than pre-ASCT PET. Detection of ctDNA in pre-ASCT plasma also appears to be predictive of relapse. In ctDNA-pos pts, given the dismal PFS, strong consideration could be given to alternative treatment strategies, e.g. CAR-T cell therapy. Furthermore, detection of ctDNA in post-ASCT plasma samples is closely associated with impending relapse, which provides an attractive platform for pre-emptive therapeutic intervention. Figure Disclosures Brown: Dynamo Therapeutics: Consultancy; Morphosys: Consultancy, Membership on an entity's Board of Directors or advisory committees, Other; Octapharma: Consultancy; Pfizer: Consultancy; Acerta: Consultancy; Sun: Research Funding; Genentech: Consultancy; Rigel Pharmaceuticals: Consultancy; Eli Lilly and Company: Consultancy; Juno/Celgene: Consultancy; Invectys: Membership on an entity's Board of Directors or advisory committees, Other: DSMC; Gilead: Consultancy, Research Funding; Astra-Zeneca: Consultancy; Janssen: Honoraria; Sunesis: Consultancy; Novartis: Consultancy; Loxo: Consultancy, Research Funding; Nextcea: Consultancy; MEI Pharma: Consultancy; Kite: Consultancy; Pharmacyclics: Consultancy; AbbVie: Consultancy; Catapult: Consultancy; BeiGene: Consultancy; Verastem: Consultancy, Research Funding; TG Therapeutics: Consultancy. Crombie:AbbVie: Research Funding; Bayer: Research Funding. Davids:Gilead Sciences: Consultancy; Zentalis: Consultancy; Sunesis: Consultancy; Syros Pharmaceuticals: Consultancy; Research to Practice: Honoraria; Merck: Consultancy; Bristol Myers Squibb: Research Funding; Janssen: Consultancy; Genentech: Consultancy, Research Funding; Eli Lilly: Consultancy; Celgene: Consultancy; AstraZeneca: Consultancy, Research Funding; BeiGene: Consultancy; Ascentage Pharma: Consultancy, Research Funding; Adaptive Biotechnologies: Consultancy; AbbVie: Consultancy; Novartis: Consultancy, Research Funding; Verastem: Consultancy, Research Funding; MEI Pharma: Consultancy, Research Funding; Surface Oncology: Research Funding; Pharmacyclics: Consultancy, Research Funding; TG Therapeutics: Consultancy, Research Funding. Fisher:Kyowa Kirin: Membership on an entity's Board of Directors or advisory committees. Jacobsen:Merck, Pharmacyclics, F. Hoffmann-LaRoche, Novartis: Research Funding; Takeda: Honoraria; Acerta, AstraZeneca, Merck: Consultancy. LaCasce:BMS: Consultancy; Research to Practice: Speakers Bureau; UptoDate: Patents & Royalties. Dahi:Kite: Consultancy. Nieto:Secura Bio: Other: Grant Support; Novartis: Other: Grant Support; Affimed: Consultancy, Other: Grant Support; Astra Zeneca: Other: Grant Support. Chen:Incyte Corporation: Consultancy; Takeda: Consultancy; Magenta: Consultancy; Kiadis: Consultancy; Actinium: Other: Data and Safety Monitoring Board Member; Equillium: Other: Data and Safety Monitoring Board Member; AbbVie: Other: Data and Safety Monitoring Board Member. Herrera:Pharmacyclics: Research Funding; Bristol Myers Squibb: Consultancy, Other: Travel, Accomodations, Expenses, Research Funding; Karyopharm: Consultancy; Merck: Consultancy, Research Funding; Genentech, Inc./F. Hoffmann-La Roche Ltd: Consultancy, Research Funding; Gilead Sciences: Consultancy, Research Funding; Seattle Genetics: Consultancy, Research Funding; Immune Design: Research Funding; AstraZeneca: Research Funding. Armand:IGM: Research Funding; Bristol-Myers Squibb: Consultancy, Honoraria, Research Funding; Affimed: Consultancy, Research Funding; ADC Therapeutics: Consultancy; Celgene: Consultancy; Pfizer: Consultancy; Infinity: Consultancy; Otsuka: Research Funding; Genentech: Research Funding; Roche: Research Funding; Tensha: Research Funding; Merck: Consultancy, Honoraria, Research Funding; Adaptive: Consultancy, Research Funding; Sigma Tau: Research Funding.

PurposeThe purpose of this paper is to construct an interval grey number NGM(1,1) direct prediction model (abbreviated as IGNGM(1,1)), which need not transform interval grey numbers sequences into real number sequences, and the Markov model is used to optimize residual sequences of IGNGM(1,1) model.Design/methodology/approachA definition equation of IGNGM(1,1) model is proposed in this paper, and its time response function is solved by recursive iteration method. Next, the optimal weight of development coefficients of two boundaries is obtained by genetic algorithm, which is designed by minimizing the average relative error based on time weighted. In addition to that, the Markov model is used to modify residual sequences.FindingsThe interval grey numbers’ sequences can be predicted directly by IGNGM(1,1) model and its residual sequences can be amended by Markov model. A case study shows that the proposed model has higher accuracy in prediction.Practical implicationsUncertainty and volatility information is widespread in practical applications, and the information can be characterized by interval grey numbers. In this paper, an interval grey numbers direct prediction model is proposed, which provides a method for predicting the uncertainty information in the real world.Originality/valueThe main contribution of this paper is to propose an IGNGM(1,1) model which can realize interval grey numbers prediction without transforming them into real number and solve the optimal weight of integral development coefficient by genetic algorithm so as to avoid the distortion of prediction results. Moreover, the Markov model is used to modify residual sequences to further improve the modeling accuracy.

Des classifications verbales associant classes de verbes avec des propriétés syntaxiques et sémantiques communes aux membres d'une classe se sont montrées utiles aussi bien dans la recherche linguistique que dans le traitement automatique des langues. Cette thèse a pour objectif de présenter des approches pour l'acquisition automatique de classes verbales pour le Français palliant ainsi partiellement le manque de ce type de ressources pour le Français. Par rapport aux classes générées, dans la plupart des approches existantes, les classes de verbes produites ne sont pas associées avec une caractérisation explicite des propriétés syntaxiques et sémantiques partagées par les membres des classes. Notre approche permet non seulement de créer des classes de verbes mais aussi d'associer ces classes avec les cadres de sous-catégorisations et les grilles thématiques partagés par les membres d'une classe. Nous utilisons deux méthodes de classification pour acquérir des classes verbales. La première est une méthode symbolique appelée \textit{Analyse Formelle de Conceptes} (FCA - Formal Concept Analysis). La deuxième exploite un algorithme de gaz neuronal croissant basé sur l'étiquetage des clusters par maximisation de vraisemblance (IGNGF - Incremental Growing Neural Gas with Feature maximisation). Pour la création des classes verbales, nous appliquons ces deux méthodes aux même ressources Françaises et Anglaises. Celle-ci sont constituées d'une part d'un lexique syntaxique pour les verbes du Français, issue de la fusion de trois ressources pour le Français existantes. D'autre part elles sont obtenues par traduction automatique en Français des classes du Verbnet anglais. Les classes verbales produites sont associées à des informations syntaxiques et sémantiques explicites sous forme de cadres de sous-catégorisations et grilles thématiques. Les classifications produites sont évaluées dans un premier temps en tant que groupements de verbes par une comparaison à une référence (proposé par \cite{SunKorhonenEtAl}). Deuxièmement, les associations aux cadres syntaxiques et aux grilles thématiques sont évaluée d'une part d'une façon intrinsèque par une comparaison à une annotation manuelle en rôles thématiques. D'autre part nous effectuons une évaluation extrinsèque en utilisant les classes verbales dans une tâche d'annotation en rôles thématiques simplifiée. Ces évaluations montrent que les classifications obtenues par les deux méthodes sont pertinentes tant par rapport aux groupement de verbes produits qu'aux associations de ces verbes avec des cadres de sous-catégorisation et des grilles thématiques. Elles présentent néanmoins des caractéristiques complémentaires. Tandis que les classes produites par FCA se sont révélées plus performantes par rapport aux associations $\langle$verbe, cadre syntaxique$\rangle$ et $\langle$verbe, grille thématique$\rangle$, les classes générées par IGNGF correspondent mieux à la classification de référence et se sont montrées plus efficaces à l'attribution de rôles thématiques.