State-of-the-art meta reinforcement learning algorithms typically assume the setting of a single agent interacting with its environment in a sequential manner. A negative side-effect of this sequential execution paradigm is that, as the environment becomes more and more challenging, and thus requiring more interaction episodes for the meta-learner, it needs the agent to reason over longer and longer time-scales. To combat the difficulty of long time-scale credit assignment, we propose an alternative parallel framework, which we name "Concurrent Meta-Reinforcement Learning" (CMRL), that transforms the temporal credit assignment problem into a multi-agent reinforcement learning one. In this multi-agent setting, a set of parallel agents are executed in the same environment and each of these "rollout" agents are given the means to communicate with each other. The goal of the communication is to coordinate, in a collaborative manner, the most efficient exploration of the shared task the agents are currently assigned. This coordination therefore represents the meta-learning aspect of the framework, as each agent can be assigned or assign itself a particular section of the current task's state space. This framework is in contrast to standard RL methods that assume that each parallel rollout occurs independently, which can potentially waste computation if many of the rollouts end up sampling the same part of the state space. Furthermore, the parallel setting enables us to define several reward sharing functions and auxiliary losses that are non-trivial to apply in the sequential setting. We demonstrate the effectiveness of our proposed CMRL at improving over sequential methods in a variety of challenging tasks.
Three years ago, we released the Omniglot dataset for developing more human-like learning algorithms. Omniglot is a one-shot learning challenge, inspired by how people can learn a new concept from just one or a few examples. Along with the dataset, we proposed a suite of five challenge tasks and a computational model based on probabilistic program induction that addresses them. The computational model, although powerful, was not meant to be the final word on Omniglot; we hoped that the machine learning community would both build on our work and develop novel approaches to tackling the challenge. In the time since, we have been pleased to see the wide adoption of Omniglot and notable technical progress. There has been genuine progress on one-shot classification, but it has been difficult to measure since researchers have adopted different splits and training procedures that make the task easier. The other four tasks, while essential components of human conceptual understanding, have received considerably less attention. We review the progress so far and conclude that neural networks are still far from human-like concept learning on Omniglot, a challenge that requires performing all of the tasks with a single model. We also discuss new tasks to stimulate further progress.
Recent efforts on training visual navigation agents conditioned on language using deep reinforcement learning have been successful in learning policies for different multimodal tasks, such as semantic goal navigation and embodied question answering. In this paper, we propose a multitask model capable of jointly learning these multimodal tasks, and transferring knowledge of words and their grounding in visual objects across the tasks. The proposed model uses a novel Dual-Attention unit to disentangle the knowledge of words in the textual representations and visual concepts in the visual representations, and align them with each other. This disentangled task-invariant alignment of representations facilitates grounding and knowledge transfer across both tasks. We show that the proposed model outperforms a range of baselines on both tasks in simulated 3D environments. We also show that this disentanglement of representations makes our model modular, interpretable, and allows for transfer to instructions containing new words by leveraging object detectors.
Transformer networks have a potential of learning longer-term dependency, but are limited by a fixed-length context in the setting of language modeling. As a solution, we propose a novel neural architecture, Transformer-XL, that enables Transformer to learn dependency beyond a fixed length without disrupting temporal coherence. Concretely, it consists of a segment-level recurrence mechanism and a novel positional encoding scheme. Our method not only enables capturing longer-term dependency, but also resolves the problem of context fragmentation. As a result, Transformer-XL learns dependency that is about 80% longer than RNNs and 450% longer than vanilla Transformers, achieves better performance on both short and long sequences, and is up to 1,800+ times faster than vanilla Transformer during evaluation. Additionally, we improve the state-of-the-art (SoTA) results of bpc/perplexity from 1.06 to 0.99 on enwiki8, from 1.13 to 1.08 on text8, from 20.5 to 18.3 on WikiText-103, from 23.7 to 21.8 on One Billion Word, and from 55.3 to 54.5 on Penn Treebank (without finetuning). Our code, pretrained models, and hyperparameters are available in both Tensorflow and PyTorch.
Sequence generation models such as recurrent networks can be trained with a diverse set of learning algorithms. For example, maximum likelihood learning is simple and efficient, yet suffers from the exposure bias problem. Reinforcement learning like policy gradient addresses the problem but can have prohibitively poor exploration efficiency. A variety of other algorithms such as RAML, SPG, and data noising, have also been developed from different perspectives. This paper establishes a formal connection between these algorithms. We present a generalized entropy regularized policy optimization formulation, and show that the apparently divergent algorithms can all be reformulated as special instances of the framework, with the only difference being the configurations of reward function and a couple of hyperparameters. The unified interpretation offers a systematic view of the varying properties of exploration and learning efficiency. Besides, based on the framework, we present a new algorithm that dynamically interpolates among the existing algorithms for improved learning. Experiments on machine translation and text summarization demonstrate the superiority of the proposed algorithm.
We study the problem of alleviating the instability issue in the GAN training procedure via new architecture design. The discrepancy between the minimax and maximin objective values could serve as a proxy for the difficulties that the alternating gradient descent encounters in the optimization of GANs. In this work, we give new results on the benefits of multi-generator architecture of GANs. We show that the minimax gap shrinks to $\epsilon$ as the number of generators increases with rate $\widetilde{O}(1/\epsilon)$. This improves over the best-known result of $\widetilde{O}(1/\epsilon^2)$. At the core of our techniques is a novel application of Shapley-Folkman lemma to the generic minimax problem, where in the literature the technique was only known to work when the objective function is restricted to the Lagrangian function of a constraint optimization problem. Our proposed Stackelberg GAN performs well experimentally in both synthetic and real-world datasets, improving Fr\'echet Inception Distance by $14.61\%$ over the previous multi-generator GANs on the benchmark datasets.
Building agents that can explore their environments intelligently is a challenging open problem. In this paper, we make a step towards understanding how a hierarchical design of the agent's policy can affect its exploration capabilities. First, we design EscapeRoom environments, where the agent must figure out how to navigate to the exit by accomplishing a number of intermediate tasks (\emph{subgoals}), such as finding keys or opening doors. Our environments are procedurally generated and vary in complexity, which can be controlled by the number of subgoals and relationships between them. Next, we propose to measure the complexity of each environment by constructing dependency graphs between the goals and analytically computing \emph{hitting times} of a random walk in the graph. We empirically evaluate Proximal Policy Optimization (PPO) with sparse and shaped rewards, a variation of policy sketches, and a hierarchical version of PPO (called HiPPO) akin to h-DQN. We show that analytically estimated \emph{hitting time} in goal dependency graphs is an informative metric of the environment complexity. We conjecture that the result should hold for environments other than navigation. Finally, we show that solving environments beyond certain level of complexity requires hierarchical approaches.
The fundamental task of general density estimation $p(x)$ has been of keen interest to machine learning. In this work, we attempt to systematically characterize methods for density estimation. Broadly speaking, most of the existing methods can be categorized into either using: \textit{a}) autoregressive models to estimate the conditional factors of the chain rule, $p(x_{i}\, |\, x_{i-1}, \ldots)$; or \textit{b}) non-linear transformations of variables of a simple base distribution. Based on the study of the characteristics of these categories, we propose multiple novel methods for each category. For example we proposed RNN based transformations to model non-Markovian dependencies. Further, through a comprehensive study over both real world and synthetic data, we show for that jointly leveraging transformations of variables and autoregressive conditional models, results in a considerable improvement in performance. We illustrate the use of our models in outlier detection and image modeling. Finally we introduce a novel data driven framework for learning a family of distributions.
Generative Adversarial Networks (GAN) can achieve promising performance on learning complex data distributions on different types of data. In this paper, we first show a straightforward extension of existing GAN algorithm is not applicable to point clouds, because the constraint required for discriminators is undefined for set data. We propose a two fold modification to GAN algorithm for learning to generate point clouds (PC-GAN). First, we combine ideas from hierarchical Bayesian modeling and implicit generative models by learning a hierarchical and interpretable sampling process. A key component of our method is that we train a posterior inference network for the hidden variables. Second, instead of using only state-of-the-art Wasserstein GAN objective, we propose a sandwiching objective, which results in a tighter Wasserstein distance estimate than the commonly used dual form. Thereby, PC-GAN defines a generic framework that can incorporate many existing GAN algorithms. We validate our claims on ModelNet40 benchmark dataset. Using the distance between generated point clouds and true meshes as metric, we find that PC-GAN trained by the sandwiching objective achieves better results on test data than the existing methods. Moreover, as a byproduct, PC- GAN learns versatile latent representations of point clouds, which can achieve competitive performance with other unsupervised learning algorithms on object recognition task. Lastly, we also provide studies on generating unseen classes of objects and transforming image to point cloud, which demonstrates the compelling generalization capability and potentials of PC-GAN.
Existing question answering (QA) datasets fail to train QA systems to perform complex reasoning and provide explanations for answers. We introduce HotpotQA, a new dataset with 113k Wikipedia-based question-answer pairs with four key features: (1) the questions require finding and reasoning over multiple supporting documents to answer; (2) the questions are diverse and not constrained to any pre-existing knowledge bases or knowledge schemas; (3) we provide sentence-level supporting facts required for reasoning, allowing QA systems to reason with strong supervision and explain the predictions; (4) we offer a new type of factoid comparison questions to test QA systems' ability to extract relevant facts and perform necessary comparison. We show that HotpotQA is challenging for the latest QA systems, and the supporting facts enable models to improve performance and make explainable predictions.