Publications

BabyAI: A Platform to Study the Sample Efficiency of Grounded Language Learning

Maxime Chevalier-Boisvert

Allowing humans to interactively train artificial agents to understand language instructions is desirable for both practical and scientific … (see more)reasons, but given the poor data efficiency of the current learning methods, this goal may require substantial research efforts. Here, we introduce the BabyAI research platform to support investigations towards including humans in the loop for grounded language learning. The BabyAI platform comprises an extensible suite of 19 levels of increasing difficulty. The levels gradually lead the agent towards acquiring a combinatorially rich synthetic language which is a proper subset of English. The platform also provides a heuristic expert agent for the purpose of simulating a human teacher. We report baseline results and estimate the amount of human involvement that would be required to train a neural network-based agent on some of the BabyAI levels. We put forward strong evidence that current deep learning methods are not yet sufficiently sample efficient when it comes to learning a language with compositional properties.

2018-12-31

ICLR.cc/2019/Conference (poster)

Break the Ceiling: Stronger Multi-scale Deep Graph Convolutional Networks

Sitao Luan

Mingde Zhao

Xiao-Wen Chang

Recently, neural network based approaches have achieved significant improvement for solving large, complex, graph-structured problems. Howev… (see more)er, their bottlenecks still need to be addressed, and the advantages of multi-scale information and deep architectures have not been sufficiently exploited. In this paper, we theoretically analyze how existing Graph Convolutional Networks (GCNs) have limited expressive power due to the constraint of the activation functions and their architectures. We generalize spectral graph convolution and deep GCN in block Krylov subspace forms and devise two architectures, both with the potential to be scaled deeper but each making use of the multi-scale information in different ways. We further show that the equivalence of these two architectures can be established under certain conditions. On several node classification tasks, with or without the help of validation, the two new architectures achieve better performance compared to many state-of-the-art methods.

2018-12-31

NeurIPS (published)

Community size effect in artificial learning systems

Olivier Tieleman

Angeliki Lazaridou

Shibl Mourad

Charles Blundell

Motivated by theories of language and communication that explain why communities with large numbers of speakers have, on average, simpler la… (see more)nguages with more regularity, we cast the representation learning problem in terms of learning to communicate . Our starting point sees the traditional autoencoder setup as a single encoder with a ﬁxed decoder partner that must learn to communicate. Generalizing from there, we introduce community -based autoencoders in which multiple encoders and decoders collectively learn representations by being randomly paired up on successive training iterations. We ﬁnd that increasing community sizes reduce idiosyncrasies in the learned codes, resulting in representations that better encode concept categories and correlate with human feature norms.

2018-12-31

ViGIL@NeurIPS (published)

dblp.uni-trier.de

Connecting Weighted Automata and Recurrent Neural Networks through Spectral Learning ( Supplementary Material ) A Proofs

More precisely, the WFA A = (α, {A}σ∈Σ,Ω) with n states and the linear 2-RNN M = (α,A,Ω) with n hidden units, where A ∈ Rn×Σ×n … (see more)is defined by A:,σ,: = A for all σ ∈ Σ, are such that fA(σ1σ2 · · ·σk) = fM (x1,x2, · · · ,xk) for all sequences of input symbols σ1, · · · , σk ∈ Σ, where for each i ∈ [k] the input vector xi ∈ RΣ is the one-hot encoding of the symbol σi. Proof. We first show by induction on k that, for any sequence σ1 · · ·σk ∈ Σ∗, the hidden state hk computed by M (see Eq. (1)) on the corresponding one-hot encoded sequence x1, · · · ,xk ∈ R satisfies hk = (A1 · · ·Ak )>α. The case k = 0 is immediate. Suppose the result true for sequences of length up to k. One can check easily check that A •2 xi = Ai for any index i. Using the induction hypothesis it then follows that hk+1 = A •1 hk •2 xk+1 = Ak+1 •1 hk = (Ak+1)hk = (Aσk+1)>(Aσ1 · · ·Ak )>α = (A1 · · ·Aσk+1)>α.

2018-12-31

(published)

www.semanticscholar.org

Data-driven Chance Constrained Programming based Electric Vehicle Penetration Analysis

Di Wu

Tracy Can Cui

Benoit Boulet

Transportation electrification has been growing rapidly in recent years. The adoption of electric vehicles (EVs) could help to release the d… (see more)ependency on oil and reduce greenhouse gas emission. However, the increasing EV adoption will also impose a high demand on the power grid and may jeopardize the grid network infrastructures. For certain high EV penetration areas, the EV charging demand may lead to transformer overloading at peak hours which makes the maximal EV penetration analysis an urgent problem to solve. This paper proposes a data-driven chance constrained programming based framework for maximal EV penetration analysis. Simulation results are presented for a real-world neighborhood level network. The proposed framework could serve as a guidance for utility companies to schedule infrastructure upgrades.

2018-12-31

(published)

www.semanticscholar.org

An Empirical Study of Batch Normalization and Group Normalization in Conditional Computation

Vincent Michalski

Vikram Voleti

Samira Ebrahimi Kahou

Anthony Ortiz

Pascal Vincent

Chris Pal

Batch normalization has been widely used to improve optimization in deep neural networks. While the uncertainty in batch statistics can act … (see more)as a regularizer, using these dataset statistics specific to the training set impairs generalization in certain tasks. Recently, alternative methods for normalizing feature activations in neural networks have been proposed. Among them, group normalization has been shown to yield similar, in some domains even superior performance to batch normalization. All these methods utilize a learned affine transformation after the normalization operation to increase representational power. Methods used in conditional computation define the parameters of these transformations as learnable functions of conditioning information. In this work, we study whether and where the conditional formulation of group normalization can improve generalization compared to conditional batch normalization. We evaluate performances on the tasks of visual question answering, few-shot learning, and conditional image generation.

2018-12-31

arXiv (preprint)

An Empirical Study of Example Forgetting During Deep Neural Network Learning

Mariya Toneva

Alessandro Sordoni

Remi Tachet des Combes

Adam Trischler

Yoshua Bengio

Geoffrey J. Gordon

Inspired by the phenomenon of catastrophic forgetting, we investigate the learning dynamics of neural networks as they train on single class… (see more)ification tasks. Our goal is to understand whether a related phenomenon occurs when data does not undergo a clear distributional shift. We define a `forgetting event' to have occurred when an individual training example transitions from being classified correctly to incorrectly over the course of learning. Across several benchmark data sets, we find that: (i) certain examples are forgotten with high frequency, and some not at all; (ii) a data set's (un)forgettable examples generalize across neural architectures; and (iii) based on forgetting dynamics, a significant fraction of examples can be omitted from the training data set while still maintaining state-of-the-art generalization performance.

2018-12-31

ICLR.cc/2019/Conference (poster)

A Geometric Perspective on Optimal Representations for Reinforcement Learning

Bellemare Marc-Emmanuel

Will Dabney

Robert Dadashi

Adrien Ali Taiga

Pablo Samuel Castro

Nicolas Roux

Dale Schuurmans

Tor Lattimore

Clare Lyle

2018-12-31

NeurIPS (published)

Gradient based sample selection for online continual learning

A continual learning agent learns online with a non-stationary and never-ending stream of data. The key to such learning process is to overc… (see more)ome the catastrophic forgetting of previously seen data, which is a well known problem of neural networks. To prevent forgetting, a replay buffer is usually employed to store the previous data for the purpose of rehearsal. Previous works often depend on task boundary and i.i.d. assumptions to properly select samples for the replay buffer. In this work, we formulate sample selection as a constraint reduction problem based on the constrained optimization view of continual learning. The goal is to select a fixed subset of constraints that best approximate the feasible region defined by the original constraints. We show that it is equivalent to maximizing the diversity of samples in the replay buffer with parameters gradient as the feature. We further develop a greedy alternative that is cheap and efficient. The advantage of the proposed method is demonstrated by comparing to other alternatives under the continual learning setting. Further comparisons are made against state of the art methods that rely on task boundaries which show comparable or even better results for our method.

2018-12-31

NeurIPS (published)

h-detach: Modifying the LSTM Gradient Towards Better Optimization

Nan Rosemary Ke

Recurrent neural networks are known for their notorious exploding and vanishing gradient problem (EVGP). This problem becomes more evident i… (see more)n tasks where the information needed to correctly solve them exist over long time scales, because EVGP prevents important gradient components from being back-propagated adequately over a large number of steps. We introduce a simple stochastic algorithm (\textit{h}-detach) that is specific to LSTM optimization and targeted towards addressing this problem. Specifically, we show that when the LSTM weights are large, the gradient components through the linear path (cell state) in the LSTM computational graph get suppressed. Based on the hypothesis that these components carry information about long term dependencies (which we show empirically), their suppression can prevent LSTMs from capturing them. Our algorithm\footnote{Our code is available at this https URL.} prevents gradients flowing through this path from getting suppressed, thus allowing the LSTM to capture such dependencies better. We show significant improvements over vanilla LSTM gradient based training in terms of convergence speed, robustness to seed and learning rate, and generalization using our modification of LSTM gradient on various benchmark datasets.

2018-12-31

ICLR.cc/2019/Conference (poster)