Filters








105 Hits in 1.1 sec

Unreproducible Research is Reproducible

Xavier Bouthillier, César Laurent, Pascal Vincent
2019 International Conference on Machine Learning  
The apparent contradiction in the title is a wordplay on the different meanings attributed to the word reproducible across different scientific fields. What we imply is that unreproducible findings can be built upon reproducible methods. Without denying the importance of facilitating the reproduction of methods, we deem important to reassert that reproduction of findings is a fundamental step of the scientific inquiry. We argue that the commendable quest towards easy deterministic
more » ... y of methods and numerical results should not have us forget the even more important necessity of ensuring the reproducibility of empirical findings and conclusions by properly accounting for essential sources of variations. We provide experiments to exemplify the brittleness of current common practice in the evaluation of models in the field of deep learning, showing that even if the results could be reproduced, a slightly different experiment would not support the findings. We hope to help clarify the distinction between exploratory and empirical research in the field of deep learning and believe more energy should be devoted to proper empirical research in our community. This work is an attempt to promote the use of more rigorous and diversified methodologies. It is not an attempt to impose a new methodology and it is not a critique on the nature of exploratory research.
dblp:conf/icml/BouthillierLV19 fatcat:a6zs5hsoufhuvmusuhkm56bqk4

Dropout as data augmentation [article]

Xavier Bouthillier, Kishore Konda, Pascal Vincent, Roland Memisevic
2016 arXiv   pre-print
Dropout is typically interpreted as bagging a large number of models sharing parameters. We show that using dropout in a network can also be interpreted as a kind of data augmentation in the input space without domain knowledge. We present an approach to projecting the dropout noise within a network back into the input space, thereby generating augmented versions of the training data, and we show that training a deterministic network on the augmented samples yields similar results. Finally, we
more » ... ropose a new dropout noise scheme based on our observations and show that it improves dropout results without adding significant computational cost.
arXiv:1506.08700v4 fatcat:mtr5qwwrirdw5o4u3ntzunf2t4

Exact gradient updates in time independent of output size for the spherical loss family [article]

Pascal Vincent, Alexandre de Brébisson, Xavier Bouthillier
2016 arXiv   pre-print
An important class of problems involves training deep neural networks with sparse prediction targets of very high dimension D. These occur naturally in e.g. neural language models or the learning of word-embeddings, often posed as predicting the probability of next words among a vocabulary of size D (e.g. 200,000). Computing the equally large, but typically non-sparse D-dimensional output vector from a last hidden layer of reasonable dimension d (e.g. 500) incurs a prohibitive O(Dd)
more » ... l cost for each example, as does updating the D × d output weight matrix and computing the gradient needed for backpropagation to previous layers. While efficient handling of large sparse network inputs is trivial, the case of large sparse targets is not, and has thus so far been sidestepped with approximate alternatives such as hierarchical softmax or sampling-based approximations during training. In this work we develop an original algorithmic approach which, for a family of loss functions that includes squared error and spherical softmax, can compute the exact loss, gradient update for the output weights, and gradient for backpropagation, all in O(d^2) per example instead of O(Dd), remarkably without ever computing the D-dimensional output. The proposed algorithm yields a speedup of up to D/4d i.e. two orders of magnitude for typical sizes, for that critical part of the computations that often dominates the training time in this kind of network architecture.
arXiv:1606.08061v1 fatcat:cjmshlm3lzbe3pl6jexawyacfu

An Evaluation of Fisher Approximations Beyond Kronecker Factorization

César Laurent, Thomas George, Xavier Bouthillier, Nicolas Ballas, Pascal Vincent
2018 International Conference on Learning Representations  
We study two coarser approximations on top of a Kronecker factorization (K-FAC) of the Fisher Information Matrix, to scale up Natural Gradient to deep and wide Convolutional Neural Networks (CNNs). The first considers the feature maps as spatially uncorrelated while the second considers only correlations among groups of channels. Both variants yield a further block-diagonal approximation tailored for CNNs, which is much more efficient to compute and invert. Experiments on the VGG11 and ResNet50
more » ... architectures show the technique can substantially speed up both K-FAC and a baseline with Batch Normalization in wall-clock time, yielding faster convergence to similar or better generalization error.
dblp:conf/iclr/LaurentGBBV18 fatcat:u3hfg4kp2ranlmmvli3hd5anyu

Fast Approximate Natural Gradient Descent in a Kronecker-factored Eigenbasis [article]

Thomas George, César Laurent, Xavier Bouthillier, Nicolas Ballas, Pascal Vincent
2021 arXiv   pre-print
Optimization algorithms that leverage gradient covariance information, such as variants of natural gradient descent (Amari, 1998), offer the prospect of yielding more effective descent directions. For models with many parameters, the covariance matrix they are based on becomes gigantic, making them inapplicable in their original form. This has motivated research into both simple diagonal approximations and more sophisticated factored approximations such as KFAC (Heskes, 2000; Martens & Grosse,
more » ... 015; Grosse & Martens, 2016). In the present work we draw inspiration from both to propose a novel approximation that is provably better than KFAC and amendable to cheap partial updates. It consists in tracking a diagonal variance, not in parameter coordinates, but in a Kronecker-factored eigenbasis, in which the diagonal approximation is likely to be more effective. Experiments show improvements over KFAC in optimization speed for several deep network architectures.
arXiv:1806.03884v2 fatcat:g3uxj6myuncjjg3ofzynp6xd4u

Bradykinin receptors: Agonists, antagonists, expression, signaling, and adaptation to sustained stimulation

François Marceau, Hélène Bachelard, Johanne Bouthillier, Jean-Philippe Fortin, Guillaume Morissette, Marie-Thérèse Bawolak, Xavier Charest-Morin, Lajos Gera
2020 International Immunopharmacology  
Bradykinin-related peptides, the kinins, are blood-derived peptides that stimulate 2 G protein-coupled receptors, the B1 and B2 receptors (B1R, B2R). The pharmacologic and molecular identities of these 2 receptor subtypes will be succinctly reviewed herein, with emphasis on drug development, receptor expression, signaling, and adaptation to persistent stimulation. Peptide and non-peptide antagonists and fluorescent ligands have been produced for each receptor. The B2R is widely and
more » ... y expressed in mammalian tissues, whereas the B1R is mostly inducible under the effect of cytokines during infection and immunopathology. The B2R is temporarily desensitized by a cycle of phosphorylation/endocytosis followed by recycling, whereas the nonphosphorylable B1R is relatively resistant to desensitization and translocated to caveolae on activation. Both receptor subtypes, mainly coupled to protein G Gq, phospholipase C and calcium signaling, mediate the vascular aspects of inflammation (vasodilation, edema formation). On this basis, icatibant, a peptide antagonist of the B2R, is approved in the management of hereditary angioedema attacks. This disease is the therapeutic showcase of the kallikrein-kinin system, with an orally bioavailable B2R antagonist under development, as well as other agents that inhibit the kinin forming protease, plasma kallikrein. Other clinical applications are still elusive despite the maturity of the medicinal chemistry efforts applied to kinin receptors.
doi:10.1016/j.intimp.2020.106305 pmid:32106060 fatcat:ym7ue4xr6ffzzpty3l4rsykvoy

Autophagic flux inhibition and lysosomogenesis ensuing cellular capture and retention of the cationic drug quinacrine in murine models

Alexandre Parks, Xavier Charest-Morin, Michael Boivin-Welch, Johanne Bouthillier, Francois Marceau
2015 PeerJ  
., 2013; Marceau, Roy & Bouthillier, 2014) .  ...  As reported for various cells of human origin (Marceau et al., 2009; Marceau, Roy & Bouthillier, 2014; Roy et al., 2013) , these cells from wild type C57BL/6 mice accumulated quinacrine under the form  ...  performed the experiments, analyzed the data. • Johanne Bouthillier performed the experiments. • Francois Marceau conceived and designed the experiments, performed the experiments, analyzed the data,  ... 
doi:10.7717/peerj.1314 pmid:26500823 pmcid:PMC4614855 fatcat:kspttnf3yfh53ifnxpb4e5yj5q

Pharmacological evidence of bradykinin regeneration from extended sequences that behave as peptidase–activated B2 receptor agonists

Xavier Charest-Morin, Caroline Roy, Émile-Jacques Fortin, Johanne Bouthillier, François Marceau
2014 Frontiers in Pharmacology  
While bradykinin (BK) is known to be degraded by angiotensin converting enzyme (ACE), we have recently discovered that Met-Lys-BK-Ser-Ser is paradoxically activated by ACE. We designed and evaluated additional "prodrug" peptides extended around the BK sequence as potential ligands that could be locally activated by vascular or blood plasma peptidases. BK regeneration was estimated using the contractility of the human umbilical vein as model of vascular functions mediated by endogenous B 2
more » ... ors (B 2 Rs) and the endocytosis of the fusion protein B 2 R-green fluorescent protein (B 2 R-GFP) expressed in Human Embryonic Kidney 293 cells. Of three BK sequences extended by a C-terminal dipeptide, BK-His-Leu had the most desirable profile, exhibiting little direct affinity for the receptor but a significant one for ACE (as shown by competition of [ 3 H]BK binding to B 2 R-GFP or of [ 3 H]enalaprilat to recombinant ACE, respectively). The potency of the contractile effect of this analog on the vein was reduced 18-fold by the ACE inhibitor enalaprilat, pharmacologically evidencing BK regeneration in situ. BK-Arg, a potential substrate of arginine carboxypeptidases, had a low affinity for B 2 Rs and its potency as a contractile agent was reduced 15-fold by tissue treatment with an inhibitor of these enzymes, Plummer's inhibitor. B 2 R-GFP internalization in response to 100 nM of the extended peptides recapitulated these findings, as enalaprilat selectively inhibited the effect of BK-His-Leu and Plummer's inhibitor, that of BK-Arg. The two peptidase inhibitors did not affect BK-induced effects in either assay. The novel C-terminally extended BKs had no or very little affinity for the kinin B 1 receptor (competition of [ 3 H]Lys-des-Arg 9 -BK binding). The feasibility of peptidase-activated B 2 R agonists is illustrated by C-terminal extensions of the BK sequence.
doi:10.3389/fphar.2014.00032 pmid:24639651 pmcid:PMC3945637 fatcat:mtt2jr35zzalnlr5ek4k5t3ai4

Efficient Exact Gradient Update for training Deep Networks with Very Large Sparse Targets [article]

Pascal Vincent, Alexandre de Brébisson, Xavier Bouthillier
2015 arXiv   pre-print
An important class of problems involves training deep neural networks with sparse prediction targets of very high dimension D. These occur naturally in e.g. neural language models or the learning of word-embeddings, often posed as predicting the probability of next words among a vocabulary of size D (e.g. 200 000). Computing the equally large, but typically non-sparse D-dimensional output vector from a last hidden layer of reasonable dimension d (e.g. 500) incurs a prohibitive O(Dd)
more » ... l cost for each example, as does updating the D x d output weight matrix and computing the gradient needed for backpropagation to previous layers. While efficient handling of large sparse network inputs is trivial, the case of large sparse targets is not, and has thus so far been sidestepped with approximate alternatives such as hierarchical softmax or sampling-based approximations during training. In this work we develop an original algorithmic approach which, for a family of loss functions that includes squared error and spherical softmax, can compute the exact loss, gradient update for the output weights, and gradient for backpropagation, all in O(d^2) per example instead of O(Dd), remarkably without ever computing the D-dimensional output. The proposed algorithm yields a speedup of D/4d , i.e. two orders of magnitude for typical sizes, for that critical part of the computations that often dominates the training time in this kind of network architecture.
arXiv:1412.7091v3 fatcat:e2iibolqnbatbm7xksqi646hrm

EmoNets: Multimodal deep learning approaches for emotion recognition in video [article]

Samira Ebrahimi Kahou, Xavier Bouthillier, Pascal Lamblin, Caglar Gulcehre, Vincent Michalski, Kishore Konda, Sébastien Jean, Pierre Froumenty, Yann Dauphin, Nicolas Boulanger-Lewandowski, Raul Chandias Ferrari, Mehdi Mirza (+5 others)
2015 arXiv   pre-print
The task of the emotion recognition in the wild (EmotiW) Challenge is to assign one of seven emotions to short video clips extracted from Hollywood style movies. The videos depict acted-out emotions under realistic conditions with a large degree of variation in attributes such as pose and illumination, making it worthwhile to explore approaches which consider combinations of features from multiple modalities for label assignment. In this paper we present our approach to learning several
more » ... st models using deep learning techniques, each focusing on one modality. Among these are a convolutional neural network, focusing on capturing visual information in detected faces, a deep belief net focusing on the representation of the audio stream, a K-Means based "bag-of-mouths" model, which extracts visual features around the mouth region and a relational autoencoder, which addresses spatio-temporal aspects of videos. We explore multiple methods for the combination of cues from these modalities into one common classifier. This achieves a considerably greater accuracy than predictions from our strongest single-modality classifier. Our method was the winning submission in the 2013 EmotiW challenge and achieved a test set accuracy of 47.67% on the 2014 dataset.
arXiv:1503.01800v2 fatcat:e2ounsvhnzctnmw5ehqobn6nwq

CD70 defines a subset of proinflammatory and CNS-pathogenic TH1/TH17 lymphocytes and is overexpressed in multiple sclerosis

Tessa Dhaeze, Laurence Tremblay, Catherine Lachance, Evelyn Peelen, Stephanie Zandee, Camille Grasmuck, Lyne Bourbonnière, Sandra Larouche, Xavier Ayrignac, Rose-Marie Rébillard, Josée Poirier, Boaz Lahav (+6 others)
2019 Cellular & Molecular Immunology  
CD70 is the unique ligand of CD27 and is expressed on immune cells only upon activation. Therefore, engagement of the costimulatory CD27/CD70 pathway is solely dependent on upregulation of CD70. However, the T cell-intrinsic effect and function of human CD70 remain underexplored. Herein, we describe that CD70 expression distinguishes proinflammatory CD4+ T lymphocytes that display an increased potential to migrate into the central nervous system (CNS). Upregulation of CD70 on CD4+ T lymphocytes
more » ... is induced by TGF-β1 and TGF-β3, which promote a pathogenic phenotype. In addition, CD70 is associated with a TH1 and TH17 profile of lymphocytes and is important for T-bet and IFN-γ expression by both T helper subtypes. Moreover, adoptive transfer of CD70-/-CD4+ T lymphocytes induced less severe experimental autoimmune encephalomyelitis (EAE) disease than transfer of WT CD4+ T lymphocytes. CD70+CD4+ T lymphocytes are found in the CNS during acute autoimmune inflammation in humans and mice, highlighting CD70 as both an immune marker and an important costimulator of highly pathogenic proinflammatory TH1/TH17 lymphocytes infiltrating the CNS.
doi:10.1038/s41423-018-0198-5 pmid:30635649 pmcid:PMC6804668 fatcat:thxph27jjvggfk4djtk5mjxqtm

Combining modality specific deep neural networks for emotion recognition in video

Samira Ebrahimi Kanou, Raul Chandias Ferrari, Mehdi Mirza, Sébastien Jean, Pierre-Luc Carrier, Yann Dauphin, Nicolas Boulanger-Lewandowski, Abhishek Aggarwal, Jeremie Zumer, Pascal Lamblin, Jean-Philippe Raymond, Christopher Pal (+15 others)
2013 Proceedings of the 15th ACM on International conference on multimodal interaction - ICMI '13  
In this paper we present the techniques used for the University of Montréal's team submissions to the 2013 Emotion Recognition in the Wild Challenge. The challenge is to classify the emotions expressed by the primary human subject in short video clips extracted from feature length movies. This involves the analysis of video clips of acted scenes lasting approximately one-two seconds, including the audio track which may contain human voices as well as background music. Our approach combines
more » ... ple deep neural networks for different data modalities, including: (1) a deep convolutional neural network for the analysis of facial expressions within video frames; (2) a deep belief net to capture audio information; (3) a deep autoencoder to model the spatiotemporal information produced by the human actions depicted within the entire scene; and (4) a shallow network architecture focused on extracted features of the mouth of the primary human subject in the scene. We discuss each of these techniques, their performance characteristics and different strategies to aggregate their predictions. Our best single model was a convolutional neural network trained to predict emotions from static frames using two large data sets, the Toronto Face Database and our own set of faces images harvested from Google image search, followed by a per frame aggregation strategy that used the challenge training data. This yielded a test set accuracy of 35.58%. Using our best strategy for aggregating our top performing models into a single predictor we were able to produce an accuracy of 41.03% on the challenge test set. These compare favorably to the challenge baseline test set accuracy of 27.56%.
doi:10.1145/2522848.2531745 dblp:conf/icmi/KanouPBFGMVCBFMJCDBAZLRDPWTSBKW13 fatcat:oejeo64eibafpbbmktje5uikci

Theano: A Python framework for fast computation of mathematical expressions [article]

The Theano Development Team: Rami Al-Rfou, Guillaume Alain, Amjad Almahairi, Christof Angermueller, Dzmitry Bahdanau, Nicolas Ballas, Frédéric Bastien, Justin Bayer, Anatoly Belikov, Alexander Belopolsky, Yoshua Bengio, Arnaud Bergeron, James Bergstra, Valentin Bisson (+90 others)
2016 arXiv   pre-print
Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. Since its introduction, it has been one of the most used CPU and GPU mathematical compilers - especially in the machine learning community - and has shown steady performance improvements. Theano is being actively and continuously developed since 2008, multiple frameworks have been built on top of it and it has been used to produce many
more » ... machine learning models. The present article is structured as follows. Section I provides an overview of the Theano software and its community. Section II presents the principal features of Theano and how to use them, and compares them with other similar projects. Section III focuses on recently-introduced functionalities and improvements. Section IV compares the performance of Theano against Torch7 and TensorFlow on several machine learning models. Section V discusses current limitations of Theano and potential ways of improving it.
arXiv:1605.02688v1 fatcat:2lcqwrk2zrbt5dyjmcofn6shhu

Économie et bibliothèques, sous la direction de Jean-Michel Salaün. Paris : Éditions du Cercle de la librairie. 1997. 234 p

France Bouthillier
1999 Documentation et bibliothèques  
France Bouthillier GSLIS Université McGill  ...  du Canada, François-Xavier Garneau, dont les quatre volumes ont paru un siècle plus tard à Québec, est fixé à 2 500$.  ... 
doi:10.7202/1032781ar fatcat:7e2ni4dmg5gvnozdeynglnq4jm

Index annuel 2003

Gilles Deschatelets
2003 Documentation et bibliothèques  
normes MPEG: de la compression a la gestion inté- grée des contenus multimédias 49(2003), n°1, p. 13-21 Partir trop jeune: Sylvie Perron (1950- 2003) Gilles Deschatelets, Gaston Bernier, France Bouthillier  ...  tuits (1898-1908) dans l'histoire de la lecture publique au Québec 49 (2003), n° 3, p. 129-135 N Normes MPEG : de la compression a la gestion intégrée des contenus multi- médias (Les) François-Xavier  ... 
doi:10.7202/1030176ar fatcat:ew4sktaa5na2dddvw5sgu2auku
« Previous Showing results 1 — 15 out of 105 results