- Lisa_teatalk - mailman.iro.umontreal.ca

No teatalk this week
by Junyoung Chung 13 Déc '16

13 Déc '16

Hi lab, just to let you guys know that there is no tea talk scheduled for this week. Due to the holiday season, our next tea talk will be in January. Best, --Junyoung

1 0

No teatalk this week
by Junyoung Chung 05 Déc '16

05 Déc '16

Hi lab, there will be no tea talk on this week due to NIPS. Best, --Junyoung

1 0

[TeaTalk & LURG] Nov. 28th, Harm de Vries on "GuessWhat?! Visual object discovery through multi-modal dialogue"
by Junyoung Chung 28 Nov '16

28 Nov '16

Hi everyone, we don't have a tea talk on this Friday (Nov 25), but instead we will have one on next Monday (Nov 28). This talk is also going to be the LURG meeting of the week. Harm will present his recent work on "GuessWhat?! Visual object discovery through multi-modal dialogue". Who: Harm de Vries When: Nov. 28th (Monday) 3PM Where: PAA3195 Title: GuessWhat?! Visual object discovery through multi-modal dialogue Abstract: We introduce GuessWhat?!, a two-player guessing game as a testbed for research on the interplay of computer vision and dialogue systems. The goal of the game is to locate an unknown object in a rich image scene by asking a sequence of questions. Higher-level image understanding, like spatial reasoning and language grounding, is required to solve the proposed task. Our key contribution is the collection of a large-scale dataset consisting of 150K human-played games with a total of 800K visual question-answer pairs on 66K images. We explain our design decisions in collecting the dataset and introduce the oracle and questioner tasks that are associated with the two players of the game. We prototyped deep learning models to establish initial baselines of the introduced tasks. Best, --Junyoung

1 1

No tea talk this week!
by Junyoung Chung 24 Nov '16

24 Nov '16

Hi everyone, as mentioned in the previous email, there will be no tea talk on Friday. We will have one on Nov 28. Best, --Junyoung

2 1

[Talk] Prof. Wulfram Gerstner on "Synaptic Plasticity to Learning Rules: an approach of theoretical neuroscience"
by Junyoung Chung 23 Nov '16

23 Nov '16

Hi everyone, on next Tuesday, we will have a guest tutorial on neuroscience by professor Wulfram Gerstner who is visiting our lab for two weeks. Looking forward to see many of you there! Who: Wulfram Gerstner When: Nov. 22 (Tuesday) 3PM-5PM Where: AA6214 Title: Synaptic Plasticity to Learning Rules: an approach of theoretical neuroscience Abstract: In this tutorial I will show what type of constraints experimental data set for biologically plausible learning rules and why inside these constraints we (as theoretical neuroscientists) have a lot of freedom to think about learning rules that are functionally relevant in learning paradigms. Taking examples from unsupervised learning, reinforcement learning, and surprise modulated learning I will point out potential links (and differences) to machine learning approaches. Bio: Wulfram Gerstner is Director of the Laboratory of Computational Neuroscience LCN at the EPFL. He studied physics at the universities of Tubingen and Munich and received a PhD from the Technical University of Munich. His research in computational neuroscience concentrates on models of spiking neurons and spike-timing dependent plasticity, on the problem of neuronal coding in single neurons and populations, as well as on the role of spatial representation for navigation of rat-like autonomous agents. He currently has a joint appointment at the School of Life Sciences and the School of Computer and Communications Sciences at the EPFL. He teaches courses for Physicists, Computer Scientists, Mathematicians, and Life Scientists. Best, --Junyoung

4 5

Dialogue-based learning talk by Jason Weston
by Tegan Maharaj 17 Nov '16

17 Nov '16

Hi everyone, The tea talk this week will be Jason Weston's talk about dialogue. Who: Jason Weston When: Nov. 17 (Today) 3:00PM Where: PAA 1207 Title: Dialog-based Language Learning Abstract: We study how to learn to use language via dialog-based language learning, where supervision is given naturally and implicitly in the response of the dialog partner during the conversation. Experiments are in two domains: the bAbI dataset of (Weston et al., 2015) and the larger movieQA dataset from (Dodge et al., 2015). We evaluate a set of baseline learning strategies on these tasks, and show that a novel model incorporating predictive lookahead is a promising approach for learning from a teacher's response. In particular, a surprising result is that it can learn to answer questions correctly without any reward-based supervision at all. This is joint work with Jiwei Li, Alex Miller, Sumit Chopra and Marc'Aurelio Ranzato Hope to see you there! Cheers, Tegan

1 0

[TeaTalk] Talk by Laurent Dinh: Density estimation using Real NVP
by Junyoung Chung 10 Nov '16

10 Nov '16

Hi everyone, on this Friday, Laurent will talk about his work "Density estimation using Real NVP". This work will be presented as an oral talk at NIPS Deep Learning Symposium. Let's give him some helpful feedbacks. Who: Laurent Dinh When: Nov. 11th (Friday) 11AM Where: AA6214 Title: Density estimation using Real NVP Abstract: Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically, designing models with tractable learning, sampling, inference and evaluation is crucial in solving this task. We extend the space of such models using real-valued non-volume preserving (real NVP) transformations, a set of powerful invertible and learnable transformations, resulting in an unsupervised learning algorithm with exact log-likelihood computation, exact sampling, exact inference of latent variables, and an interpretable latent space. We demonstrate its ability to model natural images on four datasets through sampling, log-likelihood evaluation and latent variable manipulations. Cheers, --Junyoung

1 1

No teatalk this week!
by Junyoung Chung 01 Nov '16

01 Nov '16

Hi everyone, there will be no teatalk this week due to ICLR deadline! Best, --Junyoung

1 0

Fwd: C'est celui-là, Yoshua,
by Yoshua Bengio 29 Oct '16

29 Oct '16

Please all come to this talk (in English) Wednesday 3:30pm @ 6214, by Augustin Chaintreau from Columbia. -------- Forwarded Message -------- Subject: colloque DIRO *mercredi* 2 nov, Augustin Chaintreau (Columbia) Date: Sat, 29 Oct 2016 11:32:53 -0400 From: Pierre McKenzie <mckenzie(a)iro.umontreal.ca> <mckenzie(a)iro.umontreal.ca> To: seminaires(a)iro.umontreal.ca *Fixing our Opaque, Fragmented and Disparate Big Data* par *Augustin Chaintreau * Columbia University *Mercredi 2 novembre, 15:30-16:30*, *Salle 6214*, Pavillon André-Aisenstadt Université de Montréal, 2920 Chemin de la Tour The conference will be held in English. *ATTENTION: *horaire et salle inhabituels pour un colloque du DIRO. *Résumé * Today's big data is flawed, and the threats it poses are not theoretical: We show with reproducible experiments that personalization algorithms in services used by millions pose moral hazards, that metrics of social endorsement are vastly misleading, and that the network dynamics facilitated by online interactions and sharing economies stand in the way of reducing various inequalities. Personal information collection and usage, however, ultimately bring benefits that we cannot forego, including in areas such as health, energy efficiency and public policies. Opacity, Fragmented Views, and Disparate Impact may appear embedded in the fabric of Big Data; we show, on the contrary, from multiple examples that these trends can be reversed. Our first challenge is transparency and accountability in today's personalization, for which we provide a scalable solution validated on three leading services. We then address how to leverage opportunities offered by personal data in different domains, while informing mobile consumers and social media participants on their risks. Finally, we model how parsimonious individuals disparately benefit from information shared locally on a social network. Our analysis reveals general conditions on a network spectral expansion to eventually benefit all its members, closely related to the presence to special segregations between groups of users. *Biographie* Augustin is an Assistant Professor of Computer Science at Columbia University since 2010, where he directs the Mobile Social Lab. The goal of his research is to reconcile the benefits of leveraging personal data and social networks with a commitment to fairness and privacy. His latest results address transparency in personalization, the role of human mobility in privacy across several domains, the efficiency of crowdsourced content curation, the fairness of incentives to share personal data. His research lead to 25 papers in tier-1 conferences (five receiving best or best student paper awards at ACM CoNEXT, SIGMETRICS, USENIX IMC, IEEE MASS, Algotel), covered by several media including the NYT blog, The Washington Post, the Economist, or The Guardian. An ex student of the Ecole Normale Supérieure in Paris, he earned a Ph.D in mathematics and computer science in 2006, a NSF CAREER Award in 2013 and the ACM SIGMETRICS Rising star award in 2013. He has been an active member of network and web research community, organizing the upcoming Data Transparency Lab Conference, serving in the program committees of ACM SIGMETRICS (as chair), SIGCOMM, WWW, CoNEXT (as chair), MobiCom, MobiHoc, IMC, WSDM, WWW, COSN, AAAI ICWSM, and IEEE Infocom, and as area editor for IEEE TMC, ACM SIGCOMM CCR, and ACM SIGMOBILE MC2R.

1 0

[TeaTalk] Geordie Rose on "Building a system for embodied intelligence research"
by Junyoung Chung 28 Oct '16

28 Oct '16

Hi everyone, this Friday, we will have a teatalk by Geordie Rose who has co-founded D-Wave and Kindred. Looking forward to see many of you there! Who: Geordie Rose When: Oct. 28th (Friday) 11AM Where: AA6214 Title: Building a system for embodied intelligence research Abstract: Animals, including humans, have intelligence that is embodied -- it resides within a body. Brains evolved to control these bodies, and human intelligence seems to be a side effect of this process. I would like to explore in an interactive way with the audience what characteristics a research system would have that would allow exploration of how embodied intelligence works. This research system would include robots, control systems, programming interfaces, tele-operation, and potentially other useful characteristics. Bio: Geordie Rose is co-founder and CEO of Kindred (www.kindred.ai) and was the founder of D-Wave (for more details visit gr.ai). --Junyoung

1 2

No tea talk tomorrow.
by Junyoung Chung 20 Oct '16

20 Oct '16

Hi, this is a gentle reminder for whom who might be curious about tomorrow's tea talk. There is no tea talk tomorrow! Best, --Junyoung

1 0

[TeaTalk] Puneet Mehta on The Turing Test for Commerce: Introducing msg.ai Montreal Research Lab
by Sungjin Ahn 30 Sep '16

30 Sep '16

When: Sep.30, 11am Where: AA6214 ************** The Turing Test for Commerce: Introducing msg.ai Montreal Research Lab Messaging is the top habit on the smartphone with the top 4 messaging apps having over 3 billion users. With brands becoming a part of interactive part of this experience, messaging apps are becoming the new browser with AI becoming the new user interface. This is the biggest shift in digital commerce since the smartphone. ************** About the speaker Puneet is the Founder and CEO of msg.ai. He spent much of his career as a tech executive on Wall Street, building predictive engines to power large-scale trading bots. Puneet has been named to Advertising Age’s Creativity 50 list, Business Insider’s “The Silicon Alley 100” and “35 Up-And-Coming Entrepreneurs You Need To Meet.” Puneet started his career at IBM TJ Watson Research Center and actively comments on the topics of bots, personalization and conversational ux. In his spare time, he rides and works on his vintage motorcycle, mixes music, studies minimalist industrial design, and volunteers for a non-profit. About msg.ai msg.ai is a Y-Combinator and VC-backed AI platform for conversational commerce. With $7.7mm in seed financing (not fully announced yet) from tier one VCs and some of the top AI entrepreneurs (e.g. founders of Google DeepMind, founders of OpenAI and others), msg.ai board and advisors include the last CEO of twitter and the last head of search for Google. Customers include various Fortune 500 companies that are using msg.ai platform to build messaging bots for marketing, sales and service. Consumers have an incredible appetite to interact with brands through one-on-one conversations, and our platform is the AI dialog system on all the popular messaging platforms. With offices in San Francisco, New York City, and India, msg.ai is now opening a Deep Learning research lab in Montreal, where all of company’s R&D would be performed. -- Sungjin Ahn Montreal Institute for Learning Algorithms University of Montreal Pavillon André-Aisenstadt, Office 3355 Phone: (514) 649-0296

2 1

[TeaTalk] Puneet Mehta on The Turing Test for Commerce: Introducing msg.ai Montreal Research Lab
by Sungjin Ahn 26 Sep '16

26 Sep '16

When: Sep.30, 11am Where: AA6214 ************** The Turing Test for Commerce: Introducing msg.ai Montreal Research Lab Messaging is the top habit on the smartphone with the top 4 messaging apps having over 3 billion users. With brands becoming a part of interactive part of this experience, messaging apps are becoming the new browser with AI becoming the new user interface. This is the biggest shift in digital commerce since the smartphone. ************** About the speaker Puneet is the Founder and CEO of msg.ai. He spent much of his career as a tech executive on Wall Street, building predictive engines to power large-scale trading bots. Puneet has been named to Advertising Age’s Creativity 50 list, Business Insider’s “The Silicon Alley 100” and “35 Up-And-Coming Entrepreneurs You Need To Meet.” Puneet started his career at IBM TJ Watson Research Center and actively comments on the topics of bots, personalization and conversational ux. In his spare time, he rides and works on his vintage motorcycle, mixes music, studies minimalist industrial design, and volunteers for a non-profit. About msg.ai msg.ai is a Y-Combinator and VC-backed AI platform for conversational commerce. With $7.7mm in seed financing (not fully announced yet) from tier one VCs and some of the top AI entrepreneurs (e.g. founders of Google DeepMind, founders of OpenAI and others), msg.ai board and advisors include the last CEO of twitter and the last head of search for Google. Customers include various Fortune 500 companies that are using msg.ai platform to build messaging bots for marketing, sales and service. Consumers have an incredible appetite to interact with brands through one-on-one conversations, and our platform is the AI dialog system on all the popular messaging platforms. With offices in San Francisco, New York City, and India, msg.ai is now opening a Deep Learning research lab in Montreal, where all of company’s R&D would be performed. -- Sungjin Ahn Montreal Institute for Learning Algorithms University of Montreal Pavillon André-Aisenstadt, Office 3355 Phone: (514) 649-0296

1 0

Yann Lecun
by linda p 21 Sep '16

21 Sep '16

Hi all, I would like to remind that Yann Lecun's confenrence will be held today at 15h15, room S1-151, pavillon Jean-Coutu and it will be in *english*. You're all welcome Linda Peinthière Technicienne en coordination de bureau MILA 514-343-6111 ext. 1868

1 0

Fwd: 21 septembre: journée remplie d'IA au DIRO
by Yoshua Bengio 17 Sep '16

17 Sep '16

---------- Message transféré ---------- De : Guy Lapalme <lapalme(a)iro.umontreal.ca> Date : 17 septembre 2016 à 08:56 Objet : 21 septembre: journée remplie d'IA au DIRO À : rali-olst(a)iro.umontreal.ca À tous les auditeurs de la liste RALI-OLST mercredi prochain, en plus de notre séminaire habituel à 11h30 pour lequel vous avez déjà reçu l'annonce ** Some recent research on Chinese-English Patent Machine Translation Hongzheng (Victor) Li Il y aura deux autres présentations reliées à l'IA ** L'apprentissage prédictif et le futur de l'intelligence artificielle Yann LeCun (Facebook AI Research et New York University) 15 h 15, Salle S1-151 du pavillon Jean-Coutu plus de détails à http://diro.umontreal.ca/departement/conferenciers-emerites/ ** La révolution de l’intelligence artificielle par Yoshua Bengio, Yann LeCun et Joelle Pineau (U. McGill) 18h, Amphithéâtre Ernest-Cormier (K-500) du Pavillon Roger-Gaudry Plus de détails à: http://cdlm.umontreal.ca --- Bienvenue à tous Guy Lapalme

1 0

Fwd: Conférencier émérite et récipiendaire de la médaille du 50e du DIRO, Yann LeCun
by Yoshua Bengio 16 Sep '16

16 Sep '16

Yann LeCun is coming here next Wednesday and will give a talk at 3:15pm, see below. Probably in French though. A good exercise for those who are unsure about their French comprehension. He will receive some kind of prize, too. ---------- Message transféré ---------- De : Houari Sahraoui <sahraouh(a)iro.umontreal.ca> Date : 16 septembre 2016 à 15:27 Objet : Conférencier émérite et récipiendaire de la médaille du 50e du DIRO, Yann LeCun À : seminaires(a)iro.umontreal.ca Bonjour, Vous êtes conviés à la prochaine conférence de la série "Les conférenciers émérites du 50e du DIRO". http://diro.umontreal.ca/departement/conferenciers-emerites/ Yann LeCun (Facebook AI Research et New York Univesity) Date : Mercredi 21 septembre, 15 h 15 Lieu : Salle S1-151 du pavillon Jean-Coutu Titre : L'apprentissage prédictif et le futur de l'intelligence artificielle La conférence sera suivie par la remise de la médaille du 50e anniversaire du département d'informatique et de recherche opérationnelle Résumé : Les progrès fulgurants de l'intelligence artificielle ces dernières années sont en grande partie dus à la réémergence de l'apprentissage profond et des réseaux neuronaux, conjuguée à la disponibilité de grandes quantités de données et d'ordinateurs de plus en plus rapides. Nous disposons maintenant de systèmes de reconnaissance d'images dont la précision rivalise celles des humains. Des révolutions dans des domaines tels que les voitures autonomes et l'analyse d'images médicales sont pressenties. Mais tous ces succès utilisent l'apprentissage supervisé dans lequel la machine est entraînée avec des données étiquetées par des humains. Le défi au cours des prochaines années sera de permettre aux machines d'apprendre à partir de données brutes non étiquetées, telles que des vidéos ou du texte. C'est ce qu'on appelle l'apprentissage prédictif ou non supervisé. Ce qui manque aux machines d’aujourd'hui, c'est le « sens commun », qui est enraciné dans notre compréhension des contraintes du monde physique. L'apprentissage non supervisé est vu par certains comme la clé de l'acquisition du sens commun par les machines. Biographie : Yann LeCun est directeur de Facebook AI Research (FAIR) et professeur à l’Université de New York (NYU), où il est affilié au Courant Institute of Mathematical Sciences, de même qu’au Center for Data Science. Diplômé de l’ESIEE Paris, école supérieure d’ingénieurs, en électronique et électrotechnique et de l’Université Pierre et Marie Curie, il devient chercheur aux Laboratoires Bell en 1988. M. LeCun est nommé directeur de département au AT&T Labs-Research en 1996 puis rejoint NYU en 2003 et Facebook à la fin de 2013. Ses recherches portent principalement sur l'intelligence artificielle, l'apprentissage machine, la vision artificielle et la robotique. Il est l’un des chefs de file de l’apprentissage profond (deep learning). Depuis quelques années, ce champ d’études a révolutionné l'intelligence artificielle, notamment la reconnaissance visuelle et vocale et le traitement de la langue naturelle. L’une des techniques développées par M. LeCun, le réseau convolutif, est utilisée par toutes les grandes sociétés pour la reconnaissance d'images et de la parole et pour l'assistance à la conduite des voitures. Yann LeCun a publié près de 200 articles et a reçu le Neural Network Pioneer Award ainsi que le PAMI Distinguished Researcher Award de l'IEEE. Pour l'année 2015-2016, il a été nommé titulaire de la chaire annuelle « Informatique et sciences numériques » au Collège de France.

1 0

Fwd: CFP: AISTATS 2017
by Yoshua Bengio 22 Aoû '16

22 Aoû '16

---------- Forwarded message ---------- From: Aaditya Ramdas <aramdas(a)berkeley.edu> Date: 2016-08-21 21:23 GMT-04:00 Subject: CFP: AISTATS 2017 To: Csaba Szepesvari <szepesva(a)cs.ualberta.ca>, Mark Schmidt < schmidtm(a)cs.ubc.ca>, yoshua.bengio(a)umontreal.ca Hi Csaba, Mark, Yoshua, Hope things are going well in Canada. I'm the publicity chair for AISTATS next year, with Jerry Zhu and Aarti Singh being the program chairs. We've tried to be a bit more progressive in bringing together the Stat and CS/ML/AI communities this year. Could you please help disseminate the call for papers amongst Stats, EE/CS, math, b-schools, etc in your schools and other relevant departments that do ML/Stats/AI/optimization? Thanks! AISTATS <http://aistats.org/> is an interdisciplinary gathering of researchers at the intersection of artificial intelligence, machine learning, statistics, and related areas. The 20th International Conference on Artificial Intelligence and Statistics (AISTATS <http://www.aistats.org/>) will take place in Fort Lauderdale, Florida, USA from *April 20-22, 2017*. The deadline for paper submission is *Oct 13, 2016*, with final decisions made on Jan 24, 2017. New this year: 1. *Fast-track for Electronic Journal of Statistics*: Authors of a small number of accepted papers will be invited to submit an extended version for fast-track publication in a special issue of the Electronic Journal of Statistics (EJS) after the AISTATS decisions are out. Details on how to prepare such extended journal paper submission will be announced after the AISTATS decisions. 2. *Review-sharing with NIPS*: Papers previously submitted to NIPS 2016 are required to declare their previous NIPS paper ID, and optionally supply a one-page letter of revision (similar to a revision letter to journal editors; anonymized) in supplemental materials. AISTATS reviewers will have access to the previous anonymous NIPS reviews. Other than this, all submissions will be treated equally. *Paper Submission:* Electronic submission of PDF papers is required. The main part of the paper (single PDF up to 5Mb) may be up to 8 double-column pages in length including tables/figures. References only can exceed the 8 page limit. The main part should have enough information so that reviewers are able to judge the correctness and merit of the paper. Authors may optionally submit supplementary material (up to 10Mb) as a single zip file, containing additional proofs, audio, images, video, data or source code. Reviewing any supplementary material is up to the discretion of the reviewers. *Dual Submissions Policy:* Submissions that are identical (or substantially similar) to versions that have been previously published, or accepted for publication, or that have been submitted in parallel to other conferences or journals are not appropriate for AISTATS and violate our dual submission policy. Exceptions to this rule are the following: (a) it is acceptable to submit work that has been made available as a technical report or similar, e.g., on arXiv, without citing it (to preserve anonymity). (b) Submission is permitted for papers presented or to be presented at conferences or workshops without proceedings (e.g., ICML or NIPS workshops), or with only abstracts published. The dual-submission rules apply during the whole AISTATS review period until the authors have been notified about the decision on their paper. *Double-blind review:* Papers will be selected via a rigorous double-blind peer-review process (the reviewers will not know the identities of the authors, and vice versa). It will be up to the authors to ensure the proper anonymization of their paper and supplemental materials. Violation of the above rules may lead to rejection without review. One round of author rebuttal will occur with the initial reviews available to the authors. *Evaluation Criteria:* Submissions will be judged on the basis of technical quality, novelty, potential impact, and clarity. Typical papers often (but not always) consist of a mix of algorithmic, theoretical and experimental results, in varying proportions. Results will be judged on the degree to which they have been objectively established and/or their potential for scientific and technological impact. *Publication and presentation:* All accepted papers will be presented at the conference as posters, with a few selected for additional oral presentation. All accepted papers will be treated equally when published in the AISTATS Conference Proceedings (Journal of Machine Learning Research Workshop and Conference Proceedings series). At least one author of each accepted paper must register and attend AISTATS. A small number of accepted papers will be invited to submit an extended version for fast-track publication in a special issue of the Electronic Journal of Statistics (EJS) journal after the AISTATS decisions are out. *Topics:* Since its inception in 1985, the primary goal of AISTATS has been to promote the exchange of ideas from artificial intelligence, machine learning, and statistics. We encourage the submission of all papers in keeping of this objective. Solicited topics include, but are not limited to: - Supervised, unsupervised and semi-supervised learning, kernel and Bayesian methods - Stochastic processes, hypothesis testing, causality, time-series, nonparametrics, asymptotic theory - Graphical models and inference, manifold learning and embedding, network analysis, statistical analysis of deep learning - Sparse models and compressed sensing, information theory - Reinforcement learning, planning, control, multi-agent systems, logic and probability, relational learning - Learning theory, game theoretic learning, online learning, bandits, learning for mechanism design - Convex and non-convex optimization, discrete optimization, Bayesian optimization - Algorithms and architectures for high-performance computing - Applications in biology, cognition, computer vision, natural language, neuroscience, robotics, etc. - Topological data analysis, selective inference, experimental design, interactive learning, optimal teaching, and other emerging topics --- Aaditya Ramdas www.cs.berkeley.edu/~aramdas

1 0

Fwd: [NIPS 2016] CFP "Bayesian Deep Learning" Workshop
by Yoshua Bengio 20 Aoû '16

20 Aoû '16

---------- Forwarded message ---------- From: Yarin Gal <yg279(a)cam.ac.uk> Date: 2016-08-19 4:54 GMT-04:00 Subject: [NIPS 2016] CFP "Bayesian Deep Learning" Workshop To: Yarin Gal <yg279(a)cam.ac.uk> Dear all, Could I please ask you to circulate the CFP below to people you think this might be relevant to? Thanks so much, Yarin ************************************************************* Bayesian Deep Learning workshop, NIPS 2016 Date: December 10, 2016 Location: Centre Convencions Internacional Barcelona, Barcelona, Spain http://bayesiandeeplearning.org/ ************************************************************* *1. Call for papers* We invite researchers to submit work in any of the following areas: * Probabilistic deep models for classification and regression (such as extensions and application of Bayesian neural networks), * Generative deep models (such as variational autoencoders), * Incorporating explicit prior knowledge in deep learning (such as posterior regularisation with logic rules), * Approximate inference for Bayesian deep learning (such as variational Bayes / expectation propagation / etc. in Bayesian neural networks), * Scalable MCMC inference in Bayesian deep models, * Deep recognition models for variational inference (amortised inference), * Model uncertainty in deep learning, * Bayesian deep reinforcement learning, * Deep learning with small data, * Deep learning in Bayesian modelling, * Probabilistic semi-supervised learning techniques, * Active learning and Bayesian optimisation for experimental design, * Information theory in deep learning, * Applying non-parametric methods, one-shot learning, and Bayesian deep learning in general. A submission should take the form of an extended abstract (2 pages long) in PDF format using the NIPS style. Author names do not need to be anonymised and references may extend as far as needed beyond the 2 page upper limit. If research has previously appeared in a journal, workshop, or conference (including NIPS 2016 conference), the workshop submission should extend that previous work. Submissions will be accepted as contributed talks or poster presentations. Extended abstracts should be submitted by 1 November 2016; submission details will be updated online towards the submission deadline. Final versions will be posted on the workshop website (and are archival but do not constitute a proceedings). *Key Dates:* Extended abstract submission: **1 November 2016** Acceptance notification: 16 November 2016 Travel award notification: 16 November 2016 Final paper submission: 5 December 2016 The workshop is endorsed by the International Society for Bayesian Analysis (ISBA), which will also provide a Travel Award to a graduate student or a junior researcher. *2. Description* While deep learning has been revolutionary for machine learning, most modern deep learning models cannot represent their uncertainty nor take advantage of the well studied tools of probability theory. This has started to change following recent developments of tools and techniques combining Bayesian approaches with deep learning. The intersection of the two fields has received great interest from the community over the past few years, with the introduction of new deep learning models that take advantage of Bayesian techniques, as well as Bayesian models that incorporate deep learning elements. In fact, the use of Bayesian techniques in deep learning can be traced back to the 1990s', in seminal works by Radford Neal, David MacKay, and Dayan et al.. These gave us tools to reason about deep models confidence, and achieved state-of-the-art performance on many tasks. However earlier tools did not adapt when new needs arose (such as scalability to big data), and were consequently forgotten. Such ideas are now being revisited in light of new advances in the field, yielding many exciting new results. This workshop will study the advantages and disadvantages of such ideas, and will be a platform to host the recent flourish of ideas using Bayesian approaches in deep learning and using deep learning tools in Bayesian modelling. The program will include a mix of invited talks, contributed talks, and contributed posters. Also, the historic context of key developments in the field will be explained in an invited talk, followed by a tribute talk to David MacKay's work in the field. Future directions for the field will be debated in a panel discussion. *3. Organisers* Yarin Gal (University of Cambridge) Christos Louizos (University of Amsterdam) Zoubin Ghahramani (University of Cambridge) Kevin Murphy (Google) Max Welling (University of Amsterdam)

1 0

Tea Talk today, May 6th: Junyoung Chung
by Jörg Bornschein 06 Mai '16

06 Mai '16

Hi, everyone, Junyoung Chung will talk today about recent work on Character-Level Neural Machine Translation: A Character-Level Decoder without Explicit Segmentation for Neural Machine Translation Who: Junyoung Chung When: Fri. May 6th, 14:30 Where: AA3195 Abstract: The existing machine translation systems, whether phrase-based or neural, have relied almost exclusively on word-level modelling with explicit segmentation. In this paper, we ask a fundamental question: can neural machine translation generate a character sequence without any explicit segmentation? To answer this question, we evaluate an attention-based encoder– decoder with a subword-level encoder and a character-level decoder on four language pairs–En-Cs, En-De, En-Ru and En-Fi– using the parallel corpora from WMT’15. Our experiments show that the models with a character-level decoder outperform the ones with a subword-level decoder on all of the four language pairs. Furthermore, the ensembles of neural models with a character-level decoder outperform the state-of-the-art non-neural machine translation systems on En-Cs, En-De and En-Fi and perform comparably on En-Ru. ---

1 1

Fwd: Prédoc oral, Xavier Bouthillier, lundi (9 mai) à 10h00 au 3195
by Yoshua Bengio 03 Mai '16

03 Mai '16

---------- Message transféré ---------- De : Alain Tapp <tappa(a)iro.umontreal.ca> Date : 3 mai 2016 à 14:33 Objet : Prédoc oral, Xavier Bouthillier, lundi (9 mai) à 10h00 au 3195 À : seminaires(a)iro.umontreal.ca Cc : Pascal Vincent <pascal20100(a)gmail.com> Bonjour! Vous êtes invité à la présentation de l'examen prédoctoral de Xavier Bouthillier, lundi le 9 mai à 10h00 au AA3195. Président-rapporteur: Alain Tapp Membre: Chris Pal (Ecole Polytechnique de Montréal) Directeur de Recherche: Pascal Vincent Résumé: Le traitement automatique du langage naturel est un des domaines de l’informatique aux problèmes les plus difficiles à résoudre puisqu’il est souvent impossible de les abstraire totalement du monde extérieur. Les premières tâches où l’on pu observer de bonnes performances furent des environnements spécialisés sur lesquels on appliqua des modèles à base de règles rigides. Ces modèles étaient cependant très dispendieux et très spécialisés. La puissance de généralisation de l’apprentissage machine s’est lentement intégré à ces systèmes jusqu’à finalement en remplacer totalement certains. Il reste cependant beaucoup de travail à faire avant que les problèmes que pose le traitement automatique du langage soient résolus. Dans cet ouvrage, nous abordons les problèmes de classifications de phrases, de modélisation de langage et de traduction automatique sous l’angle de l’apprentissage de représentations profondes. Premièrement, nous présentons un réseau de neurones convolutionnel inspiré des modèles n-grammes que nous appliquons à la classification de phrases. Deuxièmement, nous décrivons un algorithme permettant de réduire la complexité en temps de calcul de plusieurs ordres de grandeur de modèles tels que les modèles de langage connectionnistes. En se basant sur la même problématique, nous proposons aussi une alternative au critère de reconstruction des auto-encodeurs permettant d’éviter la reconstruction d’entrées de très grandes tailles, un problème classique en traitement automatique du langage naturelle. Finalement, nous proposons plusieurs idées pour de futurs travaux concernant la traduction automatique. Notamment, l’utilisation d’une sortie structurée permettant la construction itérative de traductions, en opposition à la construction séquentielle, et le développement d’une métrique de traduction automatique basée sur la représentation des phrases dans un espace vectoriel. -- Alain Tapp Professeur titulaire DIRO, Université de Montréal http://www.iro.umontreal.ca/~tappa/

1 0

Fwd: Soutenance de thèse: 5 mai 2016, jeudi 13 h 30
by Yoshua Bengio 29 Avr '16

29 Avr '16

Don't miss this PhD defense by Alessandro, who worked closely with me (and of course with his supervisor Jian-Yun Nie): ---------- Forwarded message ---------- From: Guy Lapalme <lapalme(a)iro.umontreal.ca> Date: 2016-04-29 10:28 GMT-04:00 Subject: Soutenance de thèse: 5 mai 2016, jeudi 13 h 30 To: seminaires(a)iro.umontreal.ca Soutenance de thèse Learning representations for information retrieval Alessandro Sordoni Le 5 mai 2016 à 13 h 30 Salle 3195, Pavillon André-Aisenstadt Information retrieval deals with questions such as: Is this document relevant to this query? How similar are two queries or two documents? How can query and document similarity be used to enhance relevance estimation? In order to answer these questions, it is necessary to create computational representations of documents and queries. Our goal is to provide new ways of estimating such representations and their relevance relationship. We present a series of research leading to the following observation: future improvements in information retrieval effectiveness has to rely on representation learning techniques instead of a manual definition of the representation space. Jury: Guy Lapalme, président Jian-Yun Nie, directeur Alain Tapp, membre Bruce Croft (University of Massachusetts, Amherst), examinateur externe

1 0

Tea Talk by Vincent Dumoulin
by Jörg Bornschein 28 Avr '16

28 Avr '16

Hi everyone, this Friday we will have Vincent Dumoulin present recent joint work with Francesco Visin: Have you accepted the transposed convolution as your personal lord and saviour? When: Friday, 29th April, 2:30pm Where: AA3195 Who: Vincent Dumoulin Abstract Have you ever struggled reading someone else's CNN code? Do you sometimes have trouble understanding why on Earth some convolutional layers produce the output size they do? Do transposed convolutional layers look like black magic to you? Then the convolution arithmetic guide Francesco Visin and I wrote (http://arxiv.org/abs/1603.07285) is just the thing you need! During this tea talk, I propose to summarize the guide with neat and colourful animations and very little math. In better terms, from the guide's abstract: We introduce a guide to help deep learning practitioners understand and manipulate convolutional neural network architectures. The guide clarifies the relationship between various properties (input shape, kernel shape, zero padding, strides and output shape) of convolutional, pooling and transposed convolutional layers, as well as the relationship between convolutional and transposed convolutional layers. Relationships are derived for various cases, and are illustrated in order to make them intuitive. See you there! j

2 1

Tea talks tomorrow, Apr 22nd, 14:30
by Jörg Bornschein 22 Avr '16

22 Avr '16

Here is another last-minute announcement for our Friday tea talk tomorrow: First we will have Mirco Ravanelli, who recently joined MILA, talk about "Some insights into distant speech recognition" (abstract below). Subsequent to the talk Harm de Vries, Sarath Chandar and Aaron Courville will present and betatest "GuessWhat?!" - their new data collection game. Don't forget to bring your laptop! PS we will bring some beers! PS2 the best player will be awarded a special prize! Where: AA3195 When: Friday, Apr. 22nd, 14:30 Abstract for Mircos talk: Some insights into distant speech recognition Building computers that understand speech represents a crucial step towards easy-to-use human-machine interfaces. Despite the significant progress made in the last decade, state-of-the-art speech recognition technologies provide a satisfactory performance only in close-talking scenarios, forcing the users to speak close to a microphone-equipped device. However, it is easy to predict that in the future people will prefer to relax the constraint of handling or wearing any device to access speech recognition services, requiring techniques able to better cope with distant-talking interactions even in adverse acoustic conditions. This talk will briefly discuss some important applications of this emerging technology and will highlight prominent issues of current systems. In the final part of the talk a new architecture solely based on machine-learning techniques, which is potentially able to better counteract the harmful effects of noise and reverberation, is proposed to overcome the major limitations of state-of-the-art technology.

2 2

Re: [Lisa_teatalk] [Lisa_labo] Tea Talk Fri. 12th, 14:30
by Yoshua Bengio 19 Avr '16

19 Avr '16

and of long term dependencies are one of my life's challenges. Le 18 avr. 2016 23:33, "Anirudh goyal" <anirudhgoyal9119(a)gmail.com> a écrit : > As you already know, I started implementing Yann Olivier's paper, I > am interested in learning too about long-term credit assignment. > [image: Web Bug from > http://t.sidekickopen04.com/e1t/o/5/f18dQhb0S7ks8dDMPbW2n0x6l2B9gXrN7sKj6v5…] > > > > > - - Anirudh > > On Mon, Apr 18, 2016 at 7:59 PM, Alex Lamb <alex6200(a)gmail.com> wrote: > >> Hello, >> >> Is anyone in the lab actively working on the RNN long term credit >> assignment problem? >> >> I would be interested in another discussion on this issue, especially if >> someone is actively working on it. >> >> Best, >> >> Alex. >> >> On Sun, Apr 17, 2016 at 8:43 PM, David Krueger < >> david.scott.krueger(a)gmail.com> wrote: >> >>> In reference to the problem of constant memory training of RNNs, I >>> mentioned the forward computation of derivatives in this talk (although I >>> didn't know the name). >>> Here is a recent paper I encountered that proposes an approximation to >>> this method; it might be interesting to anyone who's interested in this >>> problem. >>> >>> http://arxiv.org/abs/1507.07680 >>> >>> On Tue, Mar 8, 2016 at 12:16 PM, Jörg Bornschein <bornj(a)iro.umontreal.ca >>> > wrote: >>> >>>> Hi, >>>> >>>> this Friday our tea-talk will be a bit more of an brain storming >>>> session and we'll have David Krueger present and discuss some research >>>> ideas. >>>> >>>> Title: Bayesian non-parametric discriminative freeze-thaw sequential >>>> GANs, >>>> and other half-baked research ideas >>>> Who: David Krueger >>>> When: Fri. March 11th, 14:30 >>>> Where: AA 3195 >>>> >>>> >>>> >>>> Looking forward to see you there! >>>> >>>> j >>>> >>>> >>>> _______________________________________________ >>>> Lisa_labo mailing list >>>> Lisa_labo(a)iro.umontreal.ca >>>> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >>>> >>>> >>> >>> _______________________________________________ >>> Lisa_labo mailing list >>> Lisa_labo(a)iro.umontreal.ca >>> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >>> >>> >> >> _______________________________________________ >> Lisa_labo mailing list >> Lisa_labo(a)iro.umontreal.ca >> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >> >> > > _______________________________________________ > Lisa_labo mailing list > Lisa_labo(a)iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo > >

1 0

Tea Talk Fri. 12th, 14:30
by Jörg Bornschein 19 Avr '16

19 Avr '16

Hi, this Friday our tea-talk will be a bit more of an brain storming session and we'll have David Krueger present and discuss some research ideas. Title: Bayesian non-parametric discriminative freeze-thaw sequential GANs, and other half-baked research ideas Who: David Krueger When: Fri. March 11th, 14:30 Where: AA 3195 Looking forward to see you there! j

3 3