ICML 2024 论文列表

Value-based reinforcement learning is the current State-Of-The-Art due to high sampling efficiency. However, our study shows it suffers from low exploitation in early training period and bias sensitiveness. To address these issues, we propose to augment the decision-making process with hypothesis, a weak form of environment description. Our approach relies on prompting the learning agent with accurate hypotheses, and designing a ready-to-adapt policy through incremental learning. We propose the ALH algorithm, showing detailed analyses on a typical learning scheme and a diverse set of Mujoco benchmarks. Our algorithm produces a significant improvement over value-based learning algorithms and other strong baselines. Our code is available at [Github URL](https://github.com/nbtpj/ALH).

Dyna-style model-based reinforcement learning (MBRL) combines model-free agents with predictive transition models through model-based rollouts. This combination raises a critical question: “When to trust your model?”; i.e., which rollout length results in the model providing useful data? Janner et al. (2019) address this question by gradually increasing rollout lengths throughout the training. While theoretically tempting, uniform model accuracy is a fallacy that collapses at the latest when extrapolating. Instead, we propose asking the question “Where to trust your model?”. Using inherent model uncertainty to consider local accuracy, we obtain the Model-Based Actor-Critic with Uncertainty-Aware Rollout Adaption (MACURA) algorithm. We propose an easy-to-tune rollout mechanism and demonstrate substantial improvements in data efficiency and performance compared to state-of-the-art deep MBRL methods on the MuJoCo benchmark.

We propose a generalization of the classical G-optimal design concept to non-linear function classes. The criterion, termed F -design, coincides with G-design in the linear case. We compute the value of the optimal design, termed the F-condition number, for several non-linear function classes. We further provide algorithms to construct designs with a bounded F -condition number. Finally, we employ the F-design in a variety of interactive machine learning tasks, where the design is naturally useful for data collection or exploration. We show that in four diverse settings of confidence band construction, contextual bandits, model-free reinforcement learning, and active learning, F-design can be combined with existing approaches in a black-box manner to yield state-of-the-art results in known problem settings as well as to generalize to novel ones.

Asynchronous action coordination presents a pervasive challenge in Multi-Agent Systems (MAS), which can be represented as a Stackelberg game (SG). However, the scalability of existing Multi-Agent Reinforcement Learning (MARL) methods based on SG is severely restricted by network architectures or environmental settings. To address this issue, we propose the Stackelberg Decision Transformer (STEER). It efficiently manages decision-making processes by incorporating the hierarchical decision structure of SG, the modeling capability of autoregressive sequence models, and the exploratory learning methodology of MARL. Our approach exhibits broad applicability across diverse task types and environmental configurations in MAS. Experimental results demonstrate both the convergence of our method towards Stackelberg equilibrium strategies and its superiority over strong baselines in complex scenarios.

Unsupervised and self-supervised objectives, such as next token prediction, have enabled pre-training generalist models from large amounts of unlabeled data. In reinforcement learning (RL), however, finding a truly general and scalable unsupervised pre-training objective for generalist policies from offline data remains a major open question. While a number of methods have been proposed to enable generic self-supervised RL, based on principles such as goal-conditioned RL, behavioral cloning, and unsupervised skill learning, such methods remain limited in terms of either the diversity of the discovered behaviors, the need for high-quality demonstration data, or the lack of a clear adaptation mechanism for downstream tasks. In this work, we propose a novel unsupervised framework to pre-train generalist policies that capture diverse, optimal, long-horizon behaviors from unlabeled offline data such that they can be quickly adapted to any arbitrary new tasks in a zero-shot manner. Our key insight is to learn a structured representation that preserves the temporal structure of the underlying environment, and then to span this learned latent space with directional movements, which enables various zero-shot policy “prompting” schemes for downstream tasks. Through our experiments on simulated robotic locomotion and manipulation benchmarks, we show that our unsupervised policies can solve goal-conditioned and general RL tasks in a zero-shot fashion, even often outperforming prior methods designed specifically for each setting. Our code and videos are available at https://seohong.me/projects/hilp/

Motivated by the success of Transformers when applied to sequences of discrete symbols, token-based world models (TBWMs) were recently proposed as sample-efficient methods. In TBWMs, the world model consumes agent experience as a language-like sequence of tokens, where each observation constitutes a sub-sequence. However, during imagination, the sequential token-by-token generation of next observations results in a severe bottleneck, leading to long training times, poor GPU utilization, and limited representations. To resolve this bottleneck, we devise a novel Parallel Observation Prediction (POP) mechanism. POP augments a Retentive Network (RetNet) with a novel forward mode tailored to our reinforcement learning setting. We incorporate POP in a novel TBWM agent named REM (Retentive Environment Model), showcasing a 15.4x faster imagination compared to prior TBWMs. REM attains superhuman performance on 12 out of 26 games of the Atari 100K benchmark, while training in less than 12 hours. Our code is available at https://github.com/leor-c/REM

Recent works have studied *state entropy maximization* in reinforcement learning, in which the agent's objective is to learn a policy inducing high entropy over states visitation (Hazan et al., 2019). They typically assume full observability of the state of the system, so that the entropy of the observations is maximized. In practice, the agent may only get *partial* observations, e.g., a robot perceiving the state of a physical space through proximity sensors and cameras. A significant mismatch between the entropy over observations and true states of the system can arise in those settings. In this paper, we address the problem of entropy maximization over the *true states* with a decision policy conditioned on partial observations *only*. The latter is a generalization of POMDPs, which is intractable in general. We develop a memory and computationally efficient *policy gradient* method to address a first-order relaxation of the objective defined on *belief* states, providing various formal characterizations of approximation gaps, the optimization landscape, and the *hallucination* problem. This paper aims to generalize state entropy maximization to more realistic domains that meet the challenges of applications.

In unsupervised environment design, reinforcement learning agents are trained on environment configurations (levels) generated by an adversary that maximises some objective. Regret is a commonly used objective that theoretically results in a minimax regret (MMR) policy with desirable robustness guarantees; in particular, the agent's maximum regret is bounded. However, once the agent reaches this regret bound on all levels, the adversary will only sample levels where regret cannot be further reduced. Although there may be possible performance improvements to be made outside of these regret-maximising levels, learning stagnates. In this work, we introduce *Bayesian level-perfect MMR* (BLP), a refinement of the minimax regret objective that overcomes this limitation. We formally show that solving for this objective results in a subset of MMR policies, and that BLP policies act consistently with a Perfect Bayesian policy over all levels. We further introduce an algorithm, *ReMiDi*, that results in a BLP policy at convergence. We empirically demonstrate that training on levels from a minimax regret adversary causes learning to prematurely stagnate, but that ReMiDi continues learning.

Meta reinforcement learning sets a distribution over a set of tasks on which the agent can train at will, then is asked to learn an optimal policy for any test task efficiently. In this paper, we consider a finite set of tasks modeled through Markov decision processes with various dynamics. We assume to have endured a long training phase, from which the set of tasks is perfectly recovered, and we focus on regret minimization against the optimal policy in the unknown test task. Under a separation condition that states the existence of a state-action pair revealing a task against another, Chen et al. (2022) show that $O(M^2 \log(H))$ regret can be achieved, where $M, H$ are the number of tasks in the set and test episodes, respectively. In our first contribution, we demonstrate that the latter rate is nearly optimal by developing a novel lower bound for test-time regret minimization under separation, showing that a linear dependence with $M$ is unavoidable. Then, we present a family of stronger yet reasonable assumptions beyond separation, which we call strong identifiability, enabling algorithms achieving fast rates $\log (H)$ and sublinear dependence with $M$ simultaneously. Our paper provides a new understanding of the statistical barriers of test-time regret minimization and when fast rates can be achieved.

While large language models (LLMs) have achieved impressive performance in generating fluent and realistic text, controlling the generated text so that it exhibits properties such as safety, factuality, and non-toxicity remains challenging. Existing decoding-based controllable text generation methods are static in terms of the dimension of control; if the target subject is changed, they require new training. Moreover, it can quickly become prohibitive to concurrently control multiple subjects. To address these challenges, we first show that existing methods can be framed as a reinforcement learning problem, where an action-value function estimates the likelihood of a desired attribute appearing in the generated text. Then, we introduce a novel approach named SF-Gen, which leverages the concept of successor features to decouple the dynamics of LLMs from task-specific rewards. By employing successor features, our method proves to be memory-efficient and computationally efficient for both training and decoding, especially when dealing with multiple target subjects. To the best of our knowledge, our research represents the first application of successor features in text generation. In addition to its computational efficiency, the resultant language produced by our method is comparable to the SOTA (and outperforms baselines) in both control measures as well as language quality, which we demonstrate through a series of experiments in various controllable text generation tasks.

Offline reinforcement learning (RL), which seeks to learn an optimal policy using offline data, has garnered significant interest due to its potential in critical applications where online data collection is infeasible or expensive. This work explores the benefit of federated learning for offline RL, aiming at collaboratively leveraging offline datasets at multiple agents. Focusing on finite-horizon episodic tabular Markov decision processes (MDPs), we design FedLCB-Q, a variant of the popular model-free Q-learning algorithm tailored for federated offline RL. FedLCB-Q updates local Q-functions at agents with novel learning rate schedules and aggregates them at a central server using importance averaging and a carefully designed pessimistic penalty term. Our sample complexity analysis reveals that, with appropriately chosen parameters and synchronization schedules, FedLCB-Q achieves linear speedup in terms of the number of agents without requiring high-quality datasets at individual agents, as long as the local datasets collectively cover the state-action space visited by the optimal policy, highlighting the power of collaboration in the federated setting. In fact, the sample complexity almost matches that of the single-agent counterpart, as if all the data are stored at a central location, up to polynomial factors of the horizon length. Furthermore, FedLCB-Q is communication-efficient, where the number of communication rounds is only linear with respect to the horizon length up to logarithmic factors.

Sample efficiency remains a crucial challenge in applying Reinforcement Learning (RL) to real-world tasks. While recent algorithms have made significant strides in improving sample efficiency, none have achieved consistently superior performance across diverse domains. In this paper, we introduce EfficientZero V2, a general framework designed for sample-efficient RL algorithms. We have expanded the performance of EfficientZero to multiple domains, encompassing both continuous and discrete actions, as well as visual and low-dimensional inputs. With a series of improvements we propose, EfficientZero V2 outperforms the current state-of-the-art (SoTA) by a significant margin in diverse tasks under the limited data setting. EfficientZero V2 exhibits a notable advancement over the prevailing general algorithm, DreamerV3, achieving superior outcomes in 50 of 66 evaluated tasks across multiple benchmarks, including Atari 100k, Proprio Control, and Vision Control.

Multiplier design---which aims to explore a large combinatorial design space to simultaneously optimize multiple conflicting objectives---is a fundamental problem in the integrated circuits industry. Although traditional approaches tackle the multi-objective multiplier optimization problem by manually designed heuristics, reinforcement learning (RL) offers a promising approach to discover high-speed and area-efficient multipliers. However, the existing RL-based methods struggle to find Pareto-optimal circuit designs for all possible preferences, i.e., weights over objectives, in a sample-efficient manner. To address this challenge, we propose a novel hierarchical adaptive (HAVE) multi-task reinforcement learning framework. The hierarchical framework consists of a meta-agent to generate diverse multiplier preferences, and an adaptive multi-task agent to collaboratively optimize multipliers conditioned on the dynamic preferences given by the meta-agent. To the best of our knowledge, HAVE is the first to well approximate Pareto-optimal circuit designs for the entire preference space with high sample efficiency. Experiments on multipliers across a wide range of input widths demonstrate that HAVE significantly Pareto-dominates state-of-the-art approaches, achieving up to 28% larger hypervolume. Moreover, experiments demonstrate that multipliers designed by HAVE can well generalize to large-scale computation-intensive circuits.

An important challenge in the field of interpretable machine learning is to ensure that deep neural networks (DNNs) use the correct or desirable input features in performing their tasks. Concept-removal methods aim to do this by eliminating concepts that are spuriously correlated with the main task from the neural network representation of the data. However, existing methods tend to be overzealous by inadvertently removing part of the correct or desirable features as well, leading to wrong interpretations and hurting model performance. We propose an iterative algorithm that separates spurious from main-task concepts by jointly estimating two low-dimensional orthogonal subspaces of the neural network representation. By evaluating the algorithm on benchmark datasets from computer vision (Waterbirds, CelebA) and natural language processing (MultiNLI), we show it outperforms existing concept-removal methods in terms of identifying the main-task and spurious concepts, and removing only the latter.

Many settings in machine learning require the selection of a rotation representation. However, choosing a suitable representation from the many available options is challenging. This paper acts as a survey and guide through rotation representations. We walk through their properties that harm or benefit deep learning with gradient-based optimization. By consolidating insights from rotation-based learning, we provide a comprehensive overview of learning functions with rotation representations. We provide guidance on selecting representations based on whether rotations are in the model’s input or output and whether the data primarily comprises small angles.

Graph representation has gained widespread application across various machine learning domains, attributed to its ability to discern correlations among input nodes. In the realm of Multi- agent Reinforcement Learning (MARL), agents are tasked with observing other entities within their environment to determine their behavior. Conventional MARL methodologies often suffer from training difficulties if Permutation Invariant (PI) and Permutation Equivariant (PE) properties are not considered during training. The adoption of graph representation offers a solution to these challenges by conceptualizing observed entities as a graph. In this context, we introduce the Hyper Graphical Attention Policy (HGAP) Network, which employs a graph attention mechanism to fulfill the PI and PE properties, while also understanding inter-entity interactions for decision-making. HGAP is assessed across various MARL benchmarks to confirm its effectiveness and efficiency. In addition, a series of ablation studies are provided to demonstrate its adaptability, transferability, and the capability to alleviate the complexities introduced by the POMDP constraint.

A common network inference problem, arising from real-world data constraints, is how to infer a dynamic network from its time-aggregated adjacency matrix and time-varying marginals (i.e., row and column sums). Prior approaches to this problem have repurposed the classic iterative proportional fitting (IPF) procedure, also known as Sinkhorn’s algorithm, with promising empirical results. However, the statistical foundation for using IPF has not been well understood: under what settings does IPF provide principled estimation of a dynamic network from its marginals, and how well does it estimate the network? In this work, we establish such a setting, by identifying a generative network model whose maximum likelihood estimates are recovered by IPF. Our model both reveals implicit assumptions on the use of IPF in such settings and enables new analyses, such as structure-dependent error bounds on IPF’s parameter estimates. When IPF fails to converge on sparse network data, we introduce a principled algorithm that guarantees IPF converges under minimal changes to the network structure. Finally, we conduct experiments with synthetic and real-world data, which demonstrate the practical value of our theoretical and algorithmic contributions.

An important aspect in developing language models that interact with humans is aligning their behavior to be useful and unharmful for their human users. This is usually achieved by tuning the model in a way that enhances desired behaviors and inhibits undesired ones, a process referred to as alignment. In this paper, we propose a theoretical approach called Behavior Expectation Bounds (BEB) which allows us to formally investigate several inherent characteristics and limitations of alignment in large language models. Importantly, we prove that within the limits of this framework, for any behavior that has a finite probability of being exhibited by the model, there exist prompts that can trigger the model into outputting this behavior, with probability that increases with the length of the prompt. This implies that any alignment process that attenuates an undesired behavior but does not remove it altogether, is not safe against adversarial prompting attacks. Furthermore, our framework hints at the mechanism by which leading alignment approaches such as reinforcement learning from human feedback make the LLM prone to being prompted into the undesired behaviors. This theoretical result is being experimentally demonstrated in large scale by the so called contemporary "chatGPT jailbreaks", where adversarial users trick the LLM into breaking its alignment guardrails by triggering it into acting as a malicious persona. Our results expose fundamental limitations in alignment of LLMs and bring to the forefront the need to devise reliable mechanisms for ensuring AI safety.

By jointly learning multiple tasks, multi-task learning (MTL) can leverage the shared knowledge across tasks, resulting in improved data efficiency and generalization performance. However, a major challenge in MTL lies in the presence of conflicting gradients, which can hinder the fair optimization of some tasks and subsequently impede MTL's ability to achieve better overall performance. Inspired by fair resource allocation in communication networks, we formulate the optimization of MTL as a utility maximization problem, where the loss decreases across tasks are maximized under different fairness measurements. To address the problem, we propose FairGrad, a novel optimization objective. FairGrad not only enables flexible emphasis on certain tasks but also achieves a theoretical convergence guarantee. Extensive experiments demonstrate that our method can achieve state-of-the-art performance among gradient manipulation methods on a suite of multi-task benchmarks in supervised learning and reinforcement learning. Furthermore, we incorporate the idea of $\alpha$-fairness into the loss functions of various MTL methods. Extensive empirical studies demonstrate that their performance can be significantly enhanced. Code is available at https://github.com/OptMN-Lab/fairgrad.