Learning to Summarize Long Texts with Memory Compression and Transfer

We introduce Mem2Mem, a memory-to-memory mechanism for hierarchical recurrent neural network based encoder decoder architectures and we explore its use for abstractive document summarization. Mem2Mem transfers "memories" via readable/writable external memory modules that augment both the encoder and decoder. Our memory regularization compresses an encoded input article into a more compact set of sentence representations. Most importantly, the memory compression step performs implicit extraction without labels, sidestepping issues with suboptimal ground-truth data and exposure bias of hybrid extractive-abstractive summarization techniques. By allowing the decoder to read/write over the encoded input memory, the model learns to read salient information about the input article while keeping track of what has been generated. Our Mem2Mem approach yields results that are competitive with state of the art transformer based summarization methods, but with 16 times fewer parameters

Measuring Systematic Generalization in Neural Proof Generation with Transformers

We are interested in understanding how well Transformer language models (TLMs) can perform reasoning tasks when trained on knowledge encoded in the form of natural language. We investigate their systematic generalization abilities on a logical reasoning task in natural language, which involves reasoning over relationships between entities grounded in first-order logical proofs. Specifically, we perform soft theorem-proving by leveraging TLMs to generate natural language proofs. We test the generated proofs for logical consistency, along with the accuracy of the final inference. We observe length-generalization issues when evaluated on longer-than-trained sequences. However, we observe TLMs improve their generalization performance after being exposed to longer, exhaustive proofs. In addition, we discover that TLMs are able to generalize better using backward-chaining proofs compared to their forward-chaining counterparts, while they find it easier to generate forward chaining proofs. We observe that models that are not trained to generate proofs are better at generalizing to problems based on longer proofs. This suggests that Transformers have efficient internal reasoning strategies that are harder to interpret. These results highlight the systematic generalization behavior of TLMs in the context of logical reasoning, and we believe this work motivates deeper inspection of their underlying reasoning strategies.

Reinforcement Learning with Random Delays

Action and observation delays commonly occur in many Reinforcement Learning applications, such as remote control scenarios. We study the anatomy of randomly delayed environments, and show that partially resampling trajectory fragments in hindsight allows for off-policy multi-step value estimation. We apply this principle to derive Delay-Correcting Actor-Critic (DCAC), an algorithm based on Soft Actor-Critic with significantly better performance in environments with delays. This is shown theoretically and also demonstrated practically on a delay-augmented version of the MuJoCo continuous control benchmark.

Conditionally Adaptive Multi-Task Learning: Improving Transfer Learning in NLP Using Fewer Parameters & Less Data

Multi-Task Learning (MTL) has emerged as a promising approach for transferring learned knowledge across different tasks. However, multi-task learning must deal with challenges such as: overfitting to low resource tasks, catastrophic forgetting, and negative task transfer, or learning interference. Additionally, in Natural Language Processing (NLP), MTL alone has typically not reached the performance level possible through per-task fine-tuning of pretrained models. However, many fine-tuning approaches are both parameter inefficient, e.g. potentially involving one new model per task, and highly susceptible to losing knowledge acquired during pretraining. We propose a novel transformer based architecture consisting of a new conditional attention mechanism as well as a set of task conditioned modules that facilitate weight sharing. Through this construction we achieve more efficient parameter sharing and mitigate forgetting by keeping half of the weights of a pretrained model fixed. We also use a new multi-task data sampling strategy to mitigate the negative effects of data imbalance across tasks. Using this approach we are able to surpass single-task fine-tuning methods while being parameter and data efficient. With our base model, we attain 2.2% higher performance compared to a full fine-tuned BERT large model on the GLUE benchmark, adding only 5.6% more trained parameters per task (whereas naive fine-tuning potentially adds 100% of the trained parameters per task) and needing only 64.6% of the data. We show that a larger variant of our single multi-task model approach performs competitively across 26 NLP tasks and yields state-of-the-art results on a number of test and development sets.

Action-Based Representation Learning for Autonomous Driving

Human drivers produce a vast amount of data which could, in principle, be used to improve autonomous driving systems. Unfortunately, seemingly straightforward approaches for creating end-to-end driving models that map sensor data directly into driving actions are problematic in terms of interpretability, and typically have significant difficulty dealing with spurious correlations. Alternatively, we propose to use this kind of action-based driving data for learning representations. Our experiments show that an affordance-based driving model pre-trained with this approach can leverage a relatively small amount of weakly annotated imagery and outperform pure end-to-end driving models, while being more interpretable. Further, we demonstrate how this strategy outperforms previous methods based on learning inverse dynamics models as well as other methods based on heavy human supervision (ImageNet).

Adversarial Soft Advantage Fitting: Imitation Learning without Policy Optimization

Adversarial imitation learning alternates between learning a discriminator -- which tells apart expert's demonstrations from generated ones -- and a generator's policy to produce trajectories that can fool this discriminator. This alternated optimization is known to be delicate in practice since it compounds unstable adversarial training with brittle and sample-inefficient reinforcement learning. We propose to remove the burden of the policy optimization steps by leveraging a novel discriminator formulation. Specifically, our discriminator is explicitly conditioned on two policies: the one from the previous generator's iteration and a learnable policy. When optimized, this discriminator directly learns the optimal generator's policy. Consequently, our discriminator's update solves the generator's optimization problem for free: learning a policy that imitates the expert does not require an additional optimization loop. This formulation effectively cuts by half the implementation and computational burden of adversarial imitation learning algorithms by removing the reinforcement learning phase altogether. We show on a variety of tasks that our simpler approach is competitive to prevalent imitation learning methods.

AR-DAE: Towards Unbiased Neural Entropy Gradient Estimation

Entropy is ubiquitous in machine learning, but it is in general intractable to compute the entropy of the distribution of an arbitrary continuous random variable. In this paper, we propose the amortized residual denoising autoencoder (AR-DAE) to approximate the gradient of the log density function, which can be used to estimate the gradient of entropy. Amortization allows us to significantly reduce the error of the gradient approximator by approaching asymptotic optimality of a regular DAE, in which case the estimation is in theory unbiased. We conduct theoretical and experimental analyses on the approximation error of the proposed method, as well as extensive studies on heuristics to ensure its robustness. Finally, using the proposed gradient approximator to estimate the gradient of entropy, we demonstrate state-of-the-art performance on density estimation with variational autoencoders and continuous control with soft actor-critic.

On the impressive performance of randomly weighted encoders in summarization tasks

In this work, we investigate the performance of untrained randomly initialized encoders in a general class of sequence to sequence models and compare their performance with that of fully-trained encoders on the task of abstractive summarization. We hypothesize that random projections of an input text have enough representational power to encode the hierarchical structure of sentences and semantics of documents. Using a trained decoder to produce abstractive text summaries, we empirically demonstrate that architectures with untrained randomly initialized encoders perform competitively with respect to the equivalent architectures with fully-trained encoders. We further find that the capacity of the encoder not only improves overall model generalization but also closes the performance gap between untrained randomly initialized and full-trained encoders. To our knowledge, it is the first time that general sequence to sequence models with attention are assessed for trained and randomly projected representations on abstractive summarization.

Exploring Structural Inductive Biases in Emergent Communication

Human language and thought are characterized by the ability to systematically generate a potentially infinite number of complex structures (e.g., sentences) from a finite set of familiar components (e.g., words). Recent works in emergent communication have discussed the propensity of artificial agents to develop a systematically compositional language through playing co-operative referential games. The degree of structure in the input data was found to affect the compositionality of the emerged communication protocols. Thus, we explore various structural priors in multi-agent communication and propose a novel graph referential game. We compare the effect of structural inductive bias (bag-of-words, sequences and graphs) on the emergence of compositional understanding of the input concepts measured by topographic similarity and generalization to unseen combinations of familiar properties. We empirically show that graph neural networks induce a better compositional language prior and a stronger generalization to out-of-domain data. We further perform ablation studies that show the robustness of the emerged protocol in graph referential games.

Real-Time Reinforcement Learning

Markov Decision Processes (MDPs), the mathematical framework underlying most algorithms in Reinforcement Learning (RL), are often used in a way that wrongfully assumes that the state of an agent's environment does not change during action selection. As RL systems based on MDPs begin to find application in real-world safety critical situations, this mismatch between the assumptions underlying classical MDPs and the reality of real-time computation may lead to undesirable outcomes. In this paper, we introduce a new framework, in which states and actions evolve simultaneously and show how it is related to the classical MDP formulation. We analyze existing algorithms under the new real-time formulation and show why they are suboptimal when used in real-time. We then use those insights to create a new algorithm Real-Time Actor-Critic (RTAC) that outperforms the existing state-of-the-art continuous control algorithm Soft Actor-Critic both in real-time and non-real-time settings. Code and videos can be found at https://github.com/rmst/rtrl.