Yoshua Bengio

Mise à l’échelle au service du raisonnement et de l’apprentissage automatique basé sur un modèle

Scaling in the service of reasoning & model-based ML

4 avril 2023

par

Yoshua Bengio

Edward J. Hu

Une collaboration entre Mila et Relation Therapeutics pour découvrir in vitro de nouvelles associations médicamenteuses synergiques

A collaboration between Mila and Relation Therapeutics to discover novel synergistic combinations of drugs in vitro

23 mars 2022

par

Paul Bertin

Jake P. Taylor-King

Yoshua Bengio

Les réseaux de flot génératifs

15 mars 2022

par

Yoshua Bengio

Publications

Contrastive introspection (ConSpec) to rapidly identify invariant prototypes for success in RL

Chen Sun

Mila

Wannan Yang

Benjamin Alsbury-Nealy

Thomas Jiralerspong

†. BlakeRichards

Reinforcement learning (RL) algorithms have achieved notable success in recent years, but still struggle with fundamental issues in long-ter… (voir plus)m credit assignment. It remains diﬃcult to learn in situations where success is contingent upon multiple critical steps that are distant in time from each other and from a sparse reward; as is often the case in real life. Moreover, how RL algorithms assign credit in these diﬃcult situations is typically not coded in a way that can rapidly generalize to new situations. Here, we present an approach using oﬄine contrastive learning, which we call contrastive introspection (ConSpec), that can be added to any existing RL algorithm and addresses both issues. In ConSpec, a contrastive loss is used during oﬄine replay to identify invariances among successful episodes. This takes advantage of the fact that it is easier to retrospectively identify the small set of steps that success is contingent upon than it is to prospectively predict reward at every step taken in the environment. ConSpec stores this knowledge in a collection of prototypes summarizing the intermediate states required for success. During training, arrival at any state that matches these prototypes generates an intrinsic reward that is added to any external rewards. As well, the reward shaping provided by ConSpec can be made to preserve the optimal policy of the underlying RL agent. The prototypes in ConSpec provide two key beneﬁts for credit assignment: (1) They enable rapid identiﬁcation of all the critical states. (2) They do so in a readily interpretable manner, enabling out of distribution generalization when sensory features are altered. In summary, ConSpec is a modular system that can be added to any existing RL algorithm to improve its long-term credit assignment.

2021-12-31

(publié)

Discrete Factorial Representations as an Abstraction for Goal Conditioned RL

Hongyu Zang

Xin Li

Romain Laroche

Remi Tachet des Combes

2021-12-31

Neural Information Processing Systems (publié)

Discrete-Valued Neural Communication in Structured Architectures Enhances Generalization

Dianbo Liu

Alex Lamb

Kenji Kawaguchi

Anirudh Goyal

Chen Sun

Michael C. Mozer

Deep learning has advanced from fully connected architectures to structured models organized into components, e.g., the transformer composed… (voir plus) of positional elements, modular architectures divided into slots, and graph neural nets made up of nodes. In structured models, an interesting question is how to conduct dynamic and possibly sparse communication among the separate components. Here, we explore the hypothesis that restricting the transmitted information among components to discrete representations is a beneficial bottleneck. The motivating intuition is human language in which communication occurs through discrete symbols. Even though individuals have different understandings of what a "cat" is based on their specific experiences, the shared discrete token makes it possible for communication among individuals to be unimpeded by individual differences in internal representation. To discretize the values of concepts dynamically communicated among specialist components, we extend the quantization mechanism from the Vector-Quantized Variational Autoencoder to multi-headed discretization with shared codebooks and use it for discrete-valued neural communication (DVNC). Our experiments show that DVNC substantially improves systematic generalization in a variety of architectures -- transformers, modular architectures, and graph neural networks. We also show that the DVNC is robust to the choice of hyperparameters, making the method very useful in practice. Moreover, we establish a theoretical justification of our discretization process, proving that it has the ability to increase noise robustness and reduce the underlying dimensionality of the model.

2021-12-31

(publié)

Enhanced Biomedical Knowledge Discovery From Unstructured Text Using Contextual Embeddings

Iz Beltagy

Kyle Lo

Arman Cohan. 2019

Scib-500

R´ejean Ducharme

P Vincent

Rishi Bommasani

Kelly Davis

Claire Cardie

Billy Chiu

Sampo Pyysalo

Ivan Vuli´c

Extracting knowledge from large, unstruc-001 tured text corpora presents a challenge. Re-002 cently, authors have utilized unsupervised, 003… (voir plus) static word embeddings to uncover "latent 004 knowledge" contained within domain-speciﬁc 005 scientiﬁc corpora. Here semantic-similarity 006 measures between representations of concepts, 007 objects or entities were used to predict re-008 lationships, which were later veriﬁed using 009 physical methods. Static language models 010 have recently been surpassed at most down-011 stream tasks by massively pre-trained, contex-012 tual language models like BERT. Some have 013 postulated that contextualized embeddings po-014 tentially yield word representations superior 015 to static ones for knowledge-discovery pur-016 poses. In an effort to address this ques-017 tion, two biomedically-trained BERT models 018 (BioBERT, SciBERT) were used to encode 019 n = 500, 1000 or 5000 sentences containing 020 words of interest extracted from a biomedical 021 corpus (Coronavirus Open Research Dataset). 022 The n representations for the words of inter-023 est were subsequently extracted and then ag-024 gregated to yield static-equivalent word rep-025 resentations. These words belonged to the 026 vocabularies of intrinsic benchmarking tools 027 for the biomedical domain (Bio-SimVerb and 028 Bio-SimLex), which assess quality of word 029 representations using semantic-similarity and 030 relatedness measures. Using intrinsic bench-031 marking tasks, feasibility of using contextual-032 ized word representations for knowledge dis-033 covery tasks can be assessed: Word represen-034 tations that better encode described reality are 035 expected to perform better (i.e. closer to do-036 main experts). As postulated, BERT embed-037 dings outperform static counterparts

2021-12-31

(publié)

Extended Abstract Track

Amin Mansouri

Jason Hartford

Kartik Ahuja

Christian Shewmake

Simone Azeglio

Arianna Di Bernardo

Nina Miolane

2021-12-31

(publié)

S5 Framework: A Review of Self-Supervised Shared Semantic Space Optimization for Multimodal Zero-Shot Learning

Clst

Yonatan Bisk

Ari Holtzman

Jesse Thomason

Ja-740 cob

Joyce Chai

Angeliki Lapata

Jonathan Lazaridou

Alek-742 May

Nicolas sandr Nisnevich

P. PintoJoseph

Turian

Ting Chen

Simon Kornblith

Mohammad Norouzi

Yen-Chun Chen

Linjie Li

Licheng Yu

Ahmed El … (voir 89 de plus)

Faisal Kholy

Zhe Ahmed

Yu Gan

Cheng

Zihan Dai

Hanxiao Liu

Quoc V. Le

Jia Deng

Wei Dong

Richard Socher

Li-Jia Li

K. Liu

Jacob Devlin

Ming-Wei Chang

Kenton Lee

Jesse Dodge

Maarten Sap

Ana Marasovic

Gabriel Agnew

Dirk Ilharco

Groeneveld Matt

Li Dong

Nan Yang

Wenhui Wang

Furu Wei

Yang Liu

Jianfeng Wang

Ming Gao

Zhou

Xiaoyi Dong

Jia Bao

Ting Zhang

Dongdong

Weiming Chen

Lu Zhang

Dong Yuan

Fang Chen

Da-cheng Juan

Chuntian Lu

Zhen Li

Futang Peng

Aleksei Timofeev

Yi-Ting Chen

Yaxi Gao

Tom

Andrew Duerig

Tomkins Sujith

Ravi

Lukasz Kaiser

Aidan N. Gomez

Noam M. Shazeer

Niki Vaswani

Llion Parmar

Jones Jakob

Uszko-850

Alex G. Kendall

Yarin Gal

Roberto Cipolla

Salman H. Khan

Muzammal Naseer

Munawar Hayat

Waqas Zamir

Fahad Shahbaz

Khan

Ranjay Krishna

Yuke Zhu

Oliver Groth

Justin John-867

Kenji Hata

Joshua Kravitz

Stephanie Chen

Mike Lewis

Yinhan Liu

Marjan Naman Goyal

Abdelrahman Ghazvininejad

Omer Mohamed

Levy

Luke Zettlemoyer

Bohan Li

Hao Zhou

Jun-Tao He

Mingxuan Wang

Liunian Harold

Mark Li

Da Yatskar

Yin

Cho-Jui

Kai-Wei Chang

Visualbert

In this review, we aim to inspire research into 001 S elf-S upervised S hared S emantic S pace ( S5 ) 002 multimodal learning problems. We e… (voir plus)quip non-003 expert researchers with a framework of in-004 formed modeling decisions via an extensive 005 literature review, an actionable modeling check-006 list, as well as a series of novel zero-shot eval-007 uation tasks. The core idea for our S5 check-008 list lies in learning contextual multimodal in-009 teractions at various granularity levels via a 010 shared Transformer encoder with a denoising 011 loss term, which is also regularized by a con-012 trastive loss term to induce a semantic align-013 ment prior on the contextual embedding space. 014 Essentially, we aim to model human concept 015 understanding and thus learn to “put a name to 016 a face”. This ultimately enables interpretable 017 zero-shot S5 generalization on a variety of 018 novel downstream tasks. In summary, this re-019 view provides sufficient background and ac-020 tionable strategies for training cutting-edge S5 021 multimodal networks. 022

2021-12-31

(publié)

Graph-Based Active Machine Learning Method for Diverse and Novel Antimicrobial Peptides Generation and Selection

Bonaventure F. P. Dossou

Dianbo Liu

Moksh Jain

Almer M. van der Sloot

Roger Palou

Michael Tyers

As antibiotic-resistant bacterial strains are rapidly spreading worldwide, infections caused by these strains are emerging as a global crisi… (voir plus)s causing the death of millions of people every year. Antimicrobial Peptides (AMPs) are one of the candidates to tackle this problem because of their potential diversity, and ability to favorably modulate the host immune response. However, large-scale screening of new AMP candidates is expensive, time-consuming, and now affordable in developing countries, which need the treatments the most. In this work, we propose a novel active machine learning-based framework that statistically minimizes the number of wet-lab experiments needed to design new AMPs, while ensuring a high diversity and novelty of generated AMPs sequences, in multi-rounds of wet-lab AMP screening settings. Combining recurrent neural network models and a graph-based filter (GraphCC), our proposed approach delivers novel and diverse candidates and demonstrates better performances according to our defined metrics.

2021-12-31

arXiv (prépublication)

arxiv.org

Harvesting Mature Relation Extraction Models from Limited Seed Knowledge: A Self-Development Framework for DS Rule Expansion

Raphael Hoffmann

Congle Zhang

Xiao Ling

Yankai Lin

Shiqi Shen

Zhiyuan Liu

Huanbo Luan

Christopher D Manning

M. Surdeanu

John Bauer

Adriana Romero

Pietro Lio’

Xuanhui Wang

Cheng Li

Nadav Golbandi

Bendersky Marc

Najork. 2018

The

Wentao Wu … (voir 2 de plus)

Hongsong Li

Haixun Wang

Distantly-supervised relation extraction 001 (DSRE) is an effective method to scale relation 002 extraction (RE) to large unlabeled corpora … (voir plus)003 with the utilization of knowledge bases (KBs), 004 but suffers from the scale of KBs and the 005 introduced noise. 006 To alleviate the above two problems, we 007 propose a novel framework called S elf-008 devel O pment r U le ex P ansion ( SOUP ), which 009 starts from limited amount of labeled data 010 and continuously produces low-noise labels on 011 large-scaled unlabeled data by a growing learn-012 able logical rules set. 013 Specifically, SOUP achieves a mutual enhance-014 ment of RE model and logical rules set, first 015 a RE model is trained on the labeled data to 016 summarize the knowledge, then the knowledge 017 is utilized to explore candidate rules from unla-018 beled data, finally high-quality candidates are 019 selected in a graph-based ranking manner to ex-020 tend the logical rules set and new rule-labeled 021 data are provided for better RE model training. 022 Experiments on wiki20 dataset demonstrate 023 that, with limited seed knowledge from small-024 scaled manually labeled data, SOUP achieves 025 significant improvement compared to baselines 026 by producing continuous growth of both logical 027 rules and the RE model, and that labeling noise 028 of SOUP is much less than DS. Furthermore, 029 RE model enhanced by SOUP with 1.6k logical 030 rules learned from prior knowledge could pro-031 duce an equivalent performance to the model 032 trained on data labeled in DS manner by 72k 033 relational facts of KBs. 034

2021-12-31

(publié)

Is a Modular Architecture Enough?

Sarthak Mittal

Guillaume Lajoie

Inspired from human cognition, machine learning systems are gradually revealing advantages of sparser and more modular architectures. Recent… (voir plus) work demonstrates that not only do some modular architectures generalize well, but they also lead to better out-of-distribution generalization, scaling properties, learning speed, and interpretability. A key intuition behind the success of such systems is that the data generating system for most real-world settings is considered to consist of sparsely interacting parts, and endowing models with similar inductive biases will be helpful. However, the field has been lacking in a rigorous quantitative assessment of such systems because these real-world data distributions are complex and unknown. In this work, we provide a thorough assessment of common modular architectures, through the lens of simple and known modular data distributions. We highlight the benefits of modularity and sparsity and reveal insights on the challenges faced while optimizing modular systems. In doing so, we propose evaluation metrics that highlight the benefits of modularity, the regimes in which these benefits are substantial, as well as the sub-optimality of current end-to-end learned modular systems as opposed to their claimed potential.

2021-12-31

Advances in Neural Information Processing Systems 35 (NeurIPS 2022) (publié)

Neural Attentive Circuits

Nasim Rahaman

Martin Weiss

Francesco Locatello

Chris Pal

Bernhard Schölkopf

Li Erran Li

Nicolas Ballas

Recent work has seen the development of general purpose neural architectures that can be trained to perform tasks across diverse data modali… (voir plus)ties. General purpose models typically make few assumptions about the underlying data-structure and are known to perform well in the large-data regime. At the same time, there has been growing interest in modular neural architectures that represent the data using sparsely interacting modules. These models can be more robust out-of-distribution, computationally efficient, and capable of sample-efficient adaptation to new data. However, they tend to make domain-specific assumptions about the data, and present challenges in how module behavior (i.e., parameterization) and connectivity (i.e., their layout) can be jointly learned. In this work, we introduce a general purpose, yet modular neural architecture called Neural Attentive Circuits (NACs) that jointly learns the parameterization and a sparse connectivity of neural modules without using domain knowledge. NACs are best understood as the combination of two systems that are jointly trained end-to-end: one that determines the module configuration and the other that executes it on an input. We demonstrate qualitatively that NACs learn diverse and meaningful module configurations on the NLVR2 dataset without additional supervision. Quantitatively, we show that by incorporating modularity in this way, NACs improve upon a strong non-modular baseline in terms of low-shot adaptation on CIFAR and CUBs dataset by about 10%, and OOD robustness on Tiny ImageNet-R by about 2.5%. Further, we find that NACs can achieve an 8x speedup at inference time while losing less than 3% performance. Finally, we find NACs to yield competitive results on diverse data modalities spanning point-cloud classification, symbolic processing and text-classification from ASCII bytes, thereby confirming its general purpose nature.

2021-12-31

Advances in Neural Information Processing Systems 35 (NeurIPS 2022) (publié)

(Private)-Retroactive Carbon Pricing [(P)ReCaP]: A Market-based Approach for Climate Finance and Risk Assessment

Prateek Gupta

Dylan Radovic

Maarten Scholl

Andrew Williams

Christian Schroeder de Witt

Tianyu Zhang

Yang Zhang

Insufficient Social Cost of Carbon (SCC) estimation methods and short-term decision-making horizons have hindered the ability of carbon emit… (voir plus)ters to properly correct for the negative externalities of climate change, as well as the capacity of nations to balance economic and climate policy. To overcome these limitations, we introduce Retrospective Social Cost of Carbon Updating (ReSCCU), a novel mechanism that corrects for these limitations as empirically measured evidence is collected. To implement ReSCCU in the context of carbon taxation, we propose Retroactive Carbon Pricing (ReCaP), a market mechanism in which polluters offload the payment of ReSCCU adjustments to insurers. To alleviate systematic risks and minimize government involvement, we introduce the Private ReCaP (PReCaP) prediction market, which could see real-world implementation based on the engagement of a few high net-worth individuals or independent institutions.

2021-12-31

arXiv (prépublication)

arxiv.org

TaHiD: Tackling Data Hiding in Fake News Detection with News Propagation Networks

Adrien Benamira

Benjamin Devillers

Etienne Lesot

Ayush K. Ray

Manal Saadi

Fragkiskos D 587

Steven Bird

Ewan Klein

Edward Loper

Nat-593

Carlos Castillo

Marcelo Mendoza

Barbara Poblete

Daryna Dementieva

Alexander Panchenko

Jacob Devlin

Ming-Wei Chang

Kenton Lee

Ashish Vaswani

Noam M. Shazeer … (voir 8 de plus)

Niki Parmar

Adriana Romero

Pietro Lio’