ICML 2023 Papers

We consider the problem of fairly allocating a sequence of indivisible items that arrive online in an arbitrary order to a group of $n$ agents with additive normalized valuation functions, we consider the allocation of goods and chores separately and propose algorithms for approximating maximin share (MMS) allocations for both settings. When agents have identical valuation functions the problem coincides with the semi-online machine covering problem (when items are goods) and load balancing problem (when items are chores), for both of which optimal competitive ratios have been achieved. In this paper we consider the case when agents have general additive valuation functions. For the allocation of goods we show that no competitive algorithm exists even when there are only three agents and propose an optimal $0.5$-competitive algorithm for the case of two agents. For the allocation of chores we propose a $(2-1/n)$-competitive algorithm for $n\geq 3$ agents and a $\sqrt{2}\approx 1.414$-competitive algorithm for two agents. Additionally, we show that no algorithm can do better than $15/11\approx 1.364$-competitive for two agents.

This paper studies the multi-agent performative prediction (Multi-PP) games over multiplex networks. We consider a distributed learning setting where agents partially cooperate on an agent network, while during learning, the data samples drawn depend on the prediction models of the agent itself and neighboring agents on a population network. The dynamics of Multi-PP games is hence affected by the interplay between both networks. This paper concentrates on this Multi-PP game with the following contributions. Firstly, we analyze sufficient conditions for the existence of the performative stable equilibrium (PSE) and Nash equilibrium (NE) of the Multi-PP games. Secondly, we analyze the changes to the equilibrium induced by perturbed data distributions, and derive the closed-form solutions where the network topologies are explicit. Our results connect the existence of PSE/NE with strengths of agents' cooperation, and the changes of equilibrium solutions across agents with their node centrality, etc. Lastly, we show that a stochastic gradient descent (SGD) based distributed learning procedure finds the PSE under the said sufficient condition. Numerical illustrations on the network effects in Multi-PP games corroborate our findings.

In cooperative multi-agent reinforcement learning, centralized training with decentralized execution (CTDE) shows great promise for a trade-off between independent Q-learning and joint action learning. However, vanilla CTDE methods assumed a fixed number of agents could hardly adapt to real-world scenarios where dynamic team compositions typically suffer from dramatically variant partial observability. Specifically, agents with extensive sight ranges are prone to be affected by trivial environmental substrates, dubbed the "distracted attention" issue; ones with limited observation can hardly sense their teammates, degrading the cooperation quality. In this paper, we propose Complementary Attention for Multi-Agent reinforcement learning (CAMA), which applies a divide-and-conquer strategy on input entities accompanied with the complementary attention of enhancement and replenishment. Concretely, to tackle the distracted attention issue, highly contributed entities' attention is enhanced by the execution-related representation extracted via action prediction with an inverse model. For better out-of-sight-range cooperation, the lowly contributed ones are compressed to brief messages with a conditional mutual information estimator. Our CAMA facilitates stable and sustainable teamwork, which is justified by the impressive results reported on the challenging StarCraftII, MPE, and Traffic Junction benchmarks.

Executing actions in a correlated manner is a common strategy for human coordination that often leads to better cooperation, which is also potentially beneficial for cooperative multi-agent reinforcement learning (MARL). However, the recent success of MARL relies heavily on the convenient paradigm of purely decentralized execution, where there is no action correlation among agents for scalability considerations. In this work, we introduce a Bayesian network to inaugurate correlations between agents' action selections in their joint policy. Theoretically, we establish a theoretical justification for why action dependencies are beneficial by deriving the multi-agent policy gradient formula under such a Bayesian network joint policy and proving its global convergence to Nash equilibria under tabular softmax policy parameterization in cooperative Markov games. Further, by equipping existing MARL algorithms with a recent method of differentiable directed acyclic graphs (DAGs), we develop practical algorithms to learn the context-aware Bayesian network policies in scenarios with partial observability and various difficulty. We also dynamically decrease the sparsity of the learned DAG throughout the training process, which leads to weakly or even purely independent policies for decentralized execution. Empirical results on a range of MARL benchmarks show the benefits of our approach.

Value decomposition is widely used in cooperative multi-agent reinforcement learning, however, its implicit credit assignment mechanism is not yet fully understood due to black-box networks. In this work, we study an interpretable value decomposition framework via the family of generalized additive models. We present a novel method, named Neural Attention Additive Q-learning (N$\text{A}^\text{2}$Q), providing inherent intelligibility of collaboration behavior. N$\text{A}^\text{2}$Q can explicitly factorize the optimal joint policy induced by enriching shape functions to model all possible coalition of agents into individual policies. Moreover, we construct the identity semantics to promote estimating credits together with the global state and individual value functions, where local semantic masks help us diagnose whether each agent captures the relevant-task information. Extensive experiments show that N$\text{A}^\text{2}$Q consistently achieves superior performance compared to different state-of-the-art methods on all challenging tasks, while yielding human-like interpretability.

The goal of continual learning is to find a model that solves multiple learning tasks which are presented sequentially to the learner. A key challenge in this setting is that the learner may "forget" how to solve a previous task when learning a new task, a phenomenon known as catastrophic forgetting. To address this challenge, many practical methods have been proposed, including memory-based, regularization-based and expansion-based methods. However, a rigorous theoretical understanding of these methods remains elusive. This paper aims to bridge this gap between theory and practice by proposing a new continual learning framework called "Ideal Continual Learner" (ICL), which is guaranteed to avoid catastrophic forgetting by construction. We show that ICL unifies multiple well-established continual learning methods and gives new theoretical insights into the strengths and weaknesses of these methods. We also derive generalization bounds for ICL which allow us to theoretically quantify "how rehearsal affects generalization". Finally, we connect ICL to several classic subjects and research topics of modern interest, which allows us to make historical remarks and inspire future directions.

Decentralized Stochastic Gradient Descent (SGD) is an emerging neural network training approach that enables multiple agents to train a model collaboratively and simultaneously. Rather than using a central parameter server to collect gradients from all the agents, each agent keeps a copy of the model parameters and communicates with a small number of other agents to exchange model updates. Their communication, governed by the communication topology and gossip weight matrices, facilitates the exchange of model updates. The state-of-the-art approach uses the dynamic one-peer exponential-2 topology, achieving faster training times and improved scalability than the ring, grid, torus, and hypercube topologies. However, this approach requires a power-of-2 number of agents, which is impractical at scale. In this paper, we remove this restriction and propose Decentralized SGD with Communication-optimal Exact Consensus Algorithm (DSGD-CECA), which works for any number of agents while still achieving state-of-the-art properties. In particular, DSGD-CECA incurs a unit per-iteration communication overhead and an $\tilde{O}(n^3)$ transient iteration complexity. Our proof is based on newly discovered properties of gossip weight matrices and a novel approach to combine them with DSGD's convergence analysis. Numerical experiments show the efficiency of DSGD-CECA.

Multi-Agent Reinforcement Learning (MARL) has demonstrated its effectiveness in learning collaboration, but it often struggles with low-quality reward signals and high non-stationarity. In contrast, Evolutionary Algorithm (EA) has shown better convergence, robustness, and signal quality insensitivity. This paper introduces a hybrid framework, Representation Asymmetry and Collaboration Evolution (RACE), which combines EA and MARL for efficient collaboration. RACE maintains a MARL team and a population of EA teams. To enable efficient knowledge sharing and policy exploration, RACE decomposes the policies of different teams controlling the same agent into a shared nonlinear observation representation encoder and individual linear policy representations. To address the partial observation issue, we introduce Value-Aware Mutual Information Maximization to enhance the shared representation with useful information about superior global states. EA evolves the population using novel agent-level crossover and mutation operators, offering diverse experiences for MARL. Concurrently, MARL optimizes its policies and injects them into the population for evolution. The experiments on challenging continuous and discrete tasks demonstrate that RACE significantly improves the basic algorithms, consistently outperforming other algorithms. Our code is available at https://github.com/yeshenpy/RACE.

Modern deep neural networks have achieved impressive performance on tasks from image classification to natural language processing. Surprisingly, these complex systems with massive amounts of parameters exhibit the same structural properties in their last-layer features and classifiers across canonical datasets when training until convergence. In particular, it has been observed that the last-layer features collapse to their class-means, and those class-means are the vertices of a simplex Equiangular Tight Frame (ETF). This phenomenon is known as Neural Collapse (NC). Recent papers have theoretically shown that NC emerges in the global minimizers of training problems with the simplified ``unconstrained feature model''. In this context, we take a step further and prove the NC occurrences in deep linear networks for the popular mean squared error (MSE) and cross entropy (CE) losses, showing that global solutions exhibit NC properties across the linear layers. Furthermore, we extend our study to imbalanced data for MSE loss and present the first geometric analysis of NC under bias-free setting. Our results demonstrate the convergence of the last-layer features and classifiers to a geometry consisting of orthogonal vectors, whose lengths depend on the amount of data in their corresponding classes. Finally, we empirically validate our theoretical analyses on synthetic and practical network architectures with both balanced and imbalanced scenarios.

Transfer operators provide a rich framework for representing the dynamics of very general, nonlinear dynamical systems. When interacting with reproducing kernel Hilbert spaces (RKHS), descriptions of dynamics often incur prohibitive data storage requirements, motivating dataset sparsification as a precursory step to computation. Further, in practice, data is available in the form of trajectories, introducing correlation between samples. In this work, we present a method for sparse learning of transfer operators from $\beta$-mixing stochastic processes, in both discrete and continuous time, and provide sample complexity analysis extending existing theoretical guarantees for learning from non-sparse, i.i.d. data. In addressing continuous-time settings, we develop precise descriptions using covariance-type operators for the infinitesimal generator that aids in the sample complexity analysis. We empirically illustrate the efficacy of our sparse embedding approach through deterministic and stochastic nonlinear system examples.

We investigate the use of natural language to drive the generalization of policies in multi-agent settings. Unlike single-agent settings, the generalization of policies should also consider the influence of other agents. Besides, with the increasing number of entities in multi-agent settings, more agent-entity interactions are needed for language grounding, and the enormous search space could impede the learning process. Moreover, given a simple general instruction, e.g., beating all enemies, agents are required to decompose it into multiple subgoals and figure out the right one to focus on. Inspired by previous work, we try to address these issues at the entity level and propose a novel framework for language grounding in multi-agent reinforcement learning, entity divider (EnDi). EnDi enables agents to independently learn subgoal division at the entity level and act in the environment based on the associated entities. The subgoal division is regularized by agent modeling to avoid subgoal conflicts and promote coordinated strategies. Empirically, EnDi demonstrates the strong generalization ability to unseen games with new dynamics and expresses the superiority over existing methods. The code is available at https://github.com/PKU-RL/EnDi.

Many machine learning problems can be abstracted in solving game theory formulations and boil down to optimizing nested objectives, such as generative adversarial networks (GANs) and multi-agent reinforcement learning. Solving these games requires finding their stable fixed points or Nash equilibrium. However, existing algorithms for solving games suffer from empirical instability, hence demanding heavy ad-hoc tuning in practice. To tackle these challenges, we resort to the emerging scheme of Learning to Optimize (L2O), which discovers problem-specific efficient optimization algorithms through data-driven training. Our customized L2O framework for differentiable game theory problems, dubbed ``Learning to Play Games" (L2PG), seeks a stable fixed point solution, by predicting the fast update direction from the past trajectory, with a novel gradient stability-aware, sign-based loss function. We further incorporate curriculum learning and self-learning to strengthen the empirical training stability and generalization of L2PG. On test problems including quadratic games and GANs, L2PG can substantially accelerate the convergence, and demonstrates a remarkably more stable trajectory. Codes are available at https://github.com/VITA-Group/L2PG.

Understanding the impact of exploration on the behaviour of multi-agent learning has, so far, benefited from the restriction to potential, or network zero-sum games in which convergence to an equilibrium can be shown. Outside of these classes, learning dynamics rarely converge and little is known about the effect of exploration in the face of non-convergence. To progress this front, we study the smooth Q- Learning dynamics. We show that, in any network game, exploration by agents results in the convergence of Q-Learning to a neighbourhood of an equilibrium. This holds independently of whether the dynamics reach the equilibrium or display complex behaviours. We show that increasing the exploration rate decreases the size of this neighbourhood and also decreases the ability of all agents to improve their payoffs. Furthermore, in a broad class of games, the payoff performance of Q-Learning dynamics, measured by Social Welfare, decreases when the exploration rate increases. Our experiments show this to be a general phenomenon, namely that exploration leads to improved convergence of Q-Learning, at the cost of payoff performance.

We propose a tuning-free dynamic SGD step size formula, which we call Distance over Gradients (DoG). The DoG step sizes depend on simple empirical quantities (distance from the initial point and norms of gradients) and have no ``learning rate'' parameter. Theoretically, we show that, for stochastic convex optimization, a slight variation of the DoG formula enjoys strong, high-probability parameter-free convergence guarantees and iterate movement bounds. Empirically, we consider a broad range of vision and language transfer learning tasks, and show that DoG's performance is close to that of SGD with tuned learning rate. We also propose a per-layer variant of DoG that generally outperforms tuned SGD, approaching the performance of tuned Adam. A PyTorch implementation of our algorithms is available at https://github.com/formll/dog.

The mechanisms behind the success of multi-view self-supervised learning (MVSSL) are not yet fully understood. Contrastive MVSSL methods have been studied through the lens of InfoNCE, a lower bound of the Mutual Information (MI). However, the relation between other MVSSL methods and MI remains unclear. We consider a different lower bound on the MI consisting of an entropy and a reconstruction term (ER), and analyze the main MVSSL families through its lens. Through this ER bound, we show that clustering-based methods such as DeepCluster and SwAV maximize the MI. We also re-interpret the mechanisms of distillation-based approaches such as BYOL and DINO, showing that they explicitly maximize the reconstruction term and implicitly encourage a stable entropy, and we confirm this empirically. We show that replacing the objectives of common MVSSL methods with this ER bound achieves competitive performance, while making them stable when training with smaller batch sizes or smaller exponential moving average (EMA) coefficients.

In order to build artificial intelligence systems that can perceive and reason with human behavior in the real world, we must first design models that conduct complex spatio-temporal reasoning over motion sequences. Moving towards this goal, we propose the HumanMotionQA task to evaluate complex, multi-step reasoning abilities of models on long-form human motion sequences. We generate a dataset of question-answer pairs that require detecting motor cues in small portions of motion sequences, reasoning temporally about when events occur, and querying specific motion attributes. In addition, we propose NSPose, a neuro-symbolic method for this task that uses symbolic reasoning and a modular design to ground motion through learning motion concepts, attribute neural operators, and temporal relations. We demonstrate the suitability of NSPose for the HumanMotionQA task, outperforming all baseline methods.

In multi-agent reinforcement learning (MARL), independent learning (IL) often shows remarkable performance and easily scales with the number of agents. Yet, using IL can be inefficient and runs the risk of failing to successfully train, particularly in scenarios that require agents to coordinate their actions. Using centralised learning (CL) enables MARL agents to quickly learn how to coordinate their behaviour but employing CL everywhere is often prohibitively expensive in real-world applications. Besides, using CL in value-based methods often needs strong representational constraints (e.g. individual-global-max condition) that can lead to poor performance if violated. In this paper, we introduce a novel plug & play IL framework named Multi-Agent Network Selection Algorithm (MANSA) which selectively employs CL only at states that require coordination. At its core, MANSA has an additional agent that uses switching controls to quickly learn the best states to activate CL during training, using CL only where necessary and vastly reducing the computational burden of CL. Our theory proves MANSA preserves cooperative MARL convergence properties, boosts IL performance and can optimally make use of a fixed budget on the number CL calls. We show empirically in Level-based Foraging (LBF) and StarCraft Multi-agent Challenge (SMAC) that MANSA achieves fast, superior and more reliable performance while making 40% fewer CL calls in SMAC and using CL at only 1% CL calls in LBF.

Policy optimization methods with function approximation are widely used in multi-agent reinforcement learning. However, it remains elusive how to design such algorithms with statistical guarantees. Leveraging a multi-agent performance difference lemma that characterizes the landscape of multi-agent policy optimization, we find that the localized action value function serves as an ideal descent direction for each local policy. Motivated by the observation, we present a multi-agent PPO algorithm in which the local policy of each agent is updated similarly to vanilla PPO. We prove that with standard regularity conditions on the Markov game and problem-dependent quantities, our algorithm converges to the globally optimal policy at a sublinear rate. We extend our algorithm to the off-policy setting and introduce pessimism to policy evaluation, which aligns with experiments. To our knowledge, this is the first provably convergent multi-agent PPO algorithm in cooperative Markov games.

In digital online advertising, advertisers procure ad impressions simultaneously on multiple platforms, or so-called channels, such as Google Ads, Meta Ads Manager, etc., each of which consists of numerous ad auctions. We study how an advertiser maximizes total conversion (e.g. ad clicks) while satisfying aggregate return-on-investment (ROI) and budget constraints across all channels. In practice, an advertiser does not have control over, and thus cannot globally optimize, which individual ad auctions she participates in for each channel, and instead authorizes a channel to procure impressions on her behalf: the advertiser can only utilize two levers on each channel, namely setting a per-channel budget and per-channel target ROI. In this work, we first analyze the effectiveness of each of these levers for solving the advertiser's global multi-channel problem. We show that when an advertiser only optimizes over per-channel ROIs, her total conversion can be arbitrarily worse than what she could have obtained in the global problem. Further, we show that the advertiser can achieve the global optimal conversion when she only optimizes over per-channel budgets. In light of this finding, under a bandit feedback setting that mimics real-world scenarios where advertisers have limited information on ad auctions in each channels and how channels procure ads, we present an efficient learning algorithm that produces per-channel budgets whose resulting conversion approximates that of the global optimal problem.