Learning to Achieve Goals with Belief State Transformers

We introduce the "Belief State Transformer", a next-token predictor that takes both a prefix and suffix as inputs, with a novel objective of predicting both the next token for the prefix and the previous token for the suffix. The Belief State Transformer effectively learns to solve challenging problems that conventional forward-only transformers struggle with, in a domain-independent fashion. Key to this success is learning a compact belief state that captures all relevant information necessary for accurate predictions. Empirical ablations show that each component of the model is essential in difficult scenarios where standard Transformers fall short. For the task of story writing with known prefixes and suffixes, our approach outperforms the Fill-in-the-Middle method for reaching known goals and demonstrates improved performance even when the goals are unknown. Altogether, the Belief State Transformer enables more efficient goal-conditioned decoding, better test-time inference, and high-quality text representations on small scale problems.

On Limitation of Transformer for Learning HMMs

Despite the remarkable success of Transformer-based architectures in various sequential modeling tasks, such as natural language processing, computer vision, and robotics, their ability to learn basic sequential models, like Hidden Markov Models (HMMs), is still unclear. This paper investigates the performance of Transformers in learning HMMs and their variants through extensive experimentation and compares them to Recurrent Neural Networks (RNNs). We show that Transformers consistently underperform RNNs in both training speed and testing accuracy across all tested HMM models. There are even challenging HMM instances where Transformers struggle to learn, while RNNs can successfully do so. Our experiments further reveal the relation between the depth of Transformers and the longest sequence length it can effectively learn, based on the types and the complexity of HMMs. To address the limitation of transformers in modeling HMMs, we demonstrate that a variant of the Chain-of-Thought (CoT), called $\textit{block CoT}$ in the training phase, can help transformers to reduce the evaluation error and to learn longer sequences at a cost of increasing the training time. Finally, we complement our empirical findings by theoretical results proving the expressiveness of transformers in approximating HMMs with logarithmic depth.

PIAVE: A Pose-Invariant Audio-Visual Speaker Extraction Network

It is common in everyday spoken communication that we look at the turning head of a talker to listen to his/her voice. Humans see the talker to listen better, so do machines. However, previous studies on audio-visual speaker extraction have not effectively handled the varying talking face. This paper studies how to take full advantage of the varying talking face. We propose a Pose-Invariant Audio-Visual Speaker Extraction Network (PIAVE) that incorporates an additional pose-invariant view to improve audio-visual speaker extraction. Specifically, we generate the pose-invariant view from each original pose orientation, which enables the model to receive a consistent frontal view of the talker regardless of his/her head pose, therefore, forming a multi-view visual input for the speaker. Experiments on the multi-view MEAD and in-the-wild LRS3 dataset demonstrate that PIAVE outperforms the state-of-the-art and is more robust to pose variations.

Is RLHF More Difficult than Standard RL?

Reinforcement learning from Human Feedback (RLHF) learns from preference signals, while standard Reinforcement Learning (RL) directly learns from reward signals. Preferences arguably contain less information than rewards, which makes preference-based RL seemingly more difficult. This paper theoretically proves that, for a wide range of preference models, we can solve preference-based RL directly using existing algorithms and techniques for reward-based RL, with small or no extra costs. Specifically, (1) for preferences that are drawn from reward-based probabilistic models, we reduce the problem to robust reward-based RL that can tolerate small errors in rewards; (2) for general arbitrary preferences where the objective is to find the von Neumann winner, we reduce the problem to multiagent reward-based RL which finds Nash equilibria for factored Markov games under a restricted set of policies. The latter case can be further reduce to adversarial MDP when preferences only depend on the final state. We instantiate all reward-based RL subroutines by concrete provable algorithms, and apply our theory to a large class of models including tabular MDPs and MDPs with generic function approximation. We further provide guarantees when K-wise comparisons are available.

Context-lumpable stochastic bandits

We consider a contextual bandit problem with $S $ contexts and $A $ actions. In each round $t=1,2,\dots$ the learner observes a random context and chooses an action based on its past experience. The learner then observes a random reward whose mean is a function of the context and the action for the round. Under the assumption that the contexts can be lumped into $r\le \min\{S ,A \}$ groups such that the mean reward for the various actions is the same for any two contexts that are in the same group, we give an algorithm that outputs an $\epsilon$-optimal policy after using at most $\widetilde O(r (S +A )/\epsilon^2)$ samples with high probability and provide a matching $\widetilde\Omega(r (S +A )/\epsilon^2)$ lower bound. In the regret minimization setting, we give an algorithm whose cumulative regret up to time $T$ is bounded by $\widetilde O(\sqrt{r^3(S +A )T})$. To the best of our knowledge, we are the first to show the near-optimal sample complexity in the PAC setting and $\widetilde O(\sqrt{{poly}(r)(S+K)T})$ minimax regret in the online setting for this problem. We also show our algorithms can be applied to more general low-rank bandits and get improved regret bounds in some scenarios.

Optimistic Natural Policy Gradient: a Simple Efficient Policy Optimization Framework for Online RL

While policy optimization algorithms have played an important role in recent empirical success of Reinforcement Learning (RL), the existing theoretical understanding of policy optimization remains rather limited -- they are either restricted to tabular MDPs or suffer from highly suboptimal sample complexity, especial in online RL where exploration is necessary. This paper proposes a simple efficient policy optimization framework -- Optimistic NPG for online RL. Optimistic NPG can be viewed as simply combining of the classic natural policy gradient (NPG) algorithm [Kakade, 2001] with optimistic policy evaluation subroutines to encourage exploration. For $d$-dimensional linear MDPs, Optimistic NPG is computationally efficient, and learns an $\varepsilon$-optimal policy within $\tilde{O}(d^2/\varepsilon^3)$ samples, which is the first computationally efficient algorithm whose sample complexity has the optimal dimension dependence $\tilde{\Theta}(d^2)$. It also improves over state-of-the-art results of policy optimization algorithms [Zanette et al., 2021] by a factor of $d$. For general function approximation that subsumes linear MDPs, Optimistic NPG, to our best knowledge, is also the first policy optimization algorithm that achieves the polynomial sample complexity for learning near-optimal policies.

Breaking the Curse of Multiagency: Provably Efficient Decentralized Multi-Agent RL with Function Approximation

A unique challenge in Multi-Agent Reinforcement Learning (MARL) is the curse of multiagency, where the description length of the game as well as the complexity of many existing learning algorithms scale exponentially with the number of agents. While recent works successfully address this challenge under the model of tabular Markov Games, their mechanisms critically rely on the number of states being finite and small, and do not extend to practical scenarios with enormous state spaces where function approximation must be used to approximate value functions or policies. This paper presents the first line of MARL algorithms that provably resolve the curse of multiagency under function approximation. We design a new decentralized algorithm -- V-Learning with Policy Replay, which gives the first polynomial sample complexity results for learning approximate Coarse Correlated Equilibria (CCEs) of Markov Games under decentralized linear function approximation. Our algorithm always outputs Markov CCEs, and achieves an optimal rate of $\widetilde{\mathcal{O}}(\epsilon^{-2})$ for finding $\epsilon$-optimal solutions. Also, when restricted to the tabular case, our result improves over the current best decentralized result $\widetilde{\mathcal{O}}(\epsilon^{-3})$ for finding Markov CCEs. We further present an alternative algorithm -- Decentralized Optimistic Policy Mirror Descent, which finds policy-class-restricted CCEs using a polynomial number of samples. In exchange for learning a weaker version of CCEs, this algorithm applies to a wider range of problems under generic function approximation, such as linear quadratic games and MARL problems with low ''marginal'' Eluder dimension.

Optimistic MLE -- A Generic Model-based Algorithm for Partially Observable Sequential Decision Making

This paper introduces a simple efficient learning algorithms for general sequential decision making. The algorithm combines Optimism for exploration with Maximum Likelihood Estimation for model estimation, which is thus named OMLE. We prove that OMLE learns the near-optimal policies of an enormously rich class of sequential decision making problems in a polynomial number of samples. This rich class includes not only a majority of known tractable model-based Reinforcement Learning (RL) problems (such as tabular MDPs, factored MDPs, low witness rank problems, tabular weakly-revealing/observable POMDPs and multi-step decodable POMDPs), but also many new challenging RL problems especially in the partially observable setting that were not previously known to be tractable. Notably, the new problems addressed by this paper include (1) observable POMDPs with continuous observation and function approximation, where we achieve the first sample complexity that is completely independent of the size of observation space; (2) well-conditioned low-rank sequential decision making problems (also known as Predictive State Representations (PSRs)), which include and generalize all known tractable POMDP examples under a more intrinsic representation; (3) general sequential decision making problems under SAIL condition, which unifies our existing understandings of model-based RL in both fully observable and partially observable settings. SAIL condition is identified by this paper, which can be viewed as a natural generalization of Bellman/witness rank to address partial observability.

Dive into Big Model Training

The increasing scale of model size and continuous improvement of performance herald the arrival of the Big Model era. In this report, we explore what and how the big model training works by diving into training objectives and training methodologies. Specifically,training objectives describe how to leverage web-scale data to develop extremely capable and incredibly large models based on self-supervised learning, and training methodologies which are based on distributed training describe how to make big model training a reality. We summarize the existing training methodologies into three main categories: training parallelism, memory-saving technologies, and model sparsity design. Training parallelism can be categorized into data, pipeline, and tensor parallelism according to the dimension of parallelism that takes place. Memory-saving technologies are orthogonal and complementary to training parallelism. And model sparsity design further scales up the model size with a constant computational cost. A continuously updated paper list of big model training is provided at https://github.com/qhliu26/BM-Training.

A Deep Reinforcement Learning Approach for Finding Non-Exploitable Strategies in Two-Player Atari Games

This paper proposes novel, end-to-end deep reinforcement learning algorithms for learning two-player zero-sum Markov games. Our objective is to find the Nash Equilibrium policies, which are free from exploitation by adversarial opponents. Distinct from prior efforts on finding Nash equilibria in extensive-form games such as Poker, which feature tree-structured transition dynamics and discrete state space, this paper focuses on Markov games with general transition dynamics and continuous state space. We propose (1) Nash DQN algorithm, which integrates DQN with a Nash finding subroutine for the joint value functions; and (2) Nash DQN Exploiter algorithm, which additionally adopts an exploiter for guiding agent's exploration. Our algorithms are the practical variants of theoretical algorithms which are guaranteed to converge to Nash equilibria in the basic tabular setting. Experimental evaluation on both tabular examples and two-player Atari games demonstrates the robustness of the proposed algorithms against adversarial opponents, as well as their advantageous performance over existing methods.