Udari Madhushani

John P. Agapiou, Alexander Sasha Vezhnevets, Edgar A. Duéñez-Guzmán et al. · deepmind

Multi-agent artificial intelligence research promises a path to develop intelligent technologies that are more human-like and more human-compatible than those produced by "solipsistic" approaches, which do not consider interactions between agents. Melting Pot is a research tool developed to facilitate work on multi-agent artificial intelligence, and provides an evaluation protocol that measures generalization to novel social partners in a set of canonical test scenarios. Each scenario pairs a physical environment (a "substrate") with a reference set of co-players (a "background population"), to create a social situation with substantial interdependence between the individuals involved. For instance, some scenarios were inspired by institutional-economics-based accounts of natural resource management and public-good-provision dilemmas. Others were inspired by considerations from evolutionary biology, game theory, and artificial life. Melting Pot aims to cover a maximally diverse set of interdependencies and incentives. It includes the commonly-studied extreme cases of perfectly-competitive (zero-sum) motivations and perfectly-cooperative (shared-reward) motivations, but does not stop with them. As in real-life, a clear majority of scenarios in Melting Pot have mixed incentives. They are neither purely competitive nor purely cooperative and thus demand successful agents be able to navigate the resulting ambiguity. Here we describe Melting Pot 2.0, which revises and expands on Melting Pot. We also introduce support for scenarios with asymmetric roles, and explain how to integrate them into the evaluation protocol. This report also contains: (1) details of all substrates and scenarios; (2) a complete description of all baseline algorithms and results. Our intention is for it to serve as a reference for researchers using Melting Pot 2.0.

Yuchen Xiao, Yanchao Sun, Mengda Xu et al.

Recent advancements in large language models (LLMs) have exhibited promising performance in solving sequential decision-making problems. By imitating few-shot examples provided in the prompts (i.e., in-context learning), an LLM agent can interact with an external environment and complete given tasks without additional training. However, such few-shot examples are often insufficient to generate high-quality solutions for complex and long-horizon tasks, while the limited context length cannot consume larger-scale demonstrations with long interaction horizons. To this end, we propose an offline learning framework that utilizes offline data at scale (e.g, logs of human interactions) to improve LLM-powered policies without finetuning. The proposed method O3D (Offline Data-driven Discovery and Distillation) automatically discovers reusable skills and distills generalizable knowledge across multiple tasks based on offline interaction data, advancing the capability of solving downstream tasks. Empirical results under two interactive decision-making benchmarks (ALFWorld and WebShop) verify that O3D can notably enhance the decision-making capabilities of LLMs through the offline discovery and distillation process, and consistently outperform baselines across various LLMs.

Udaya Ghai, Udari Madhushani, Naomi Leonard et al.

We study the problem of multi-agent control of a dynamical system with known dynamics and adversarial disturbances. Our study focuses on optimal control without centralized precomputed policies, but rather with adaptive control policies for the different agents that are only equipped with a stabilizing controller. We give a reduction from any (standard) regret minimizing control method to a distributed algorithm. The reduction guarantees that the resulting distributed algorithm has low regret relative to the optimal precomputed joint policy. Our methodology involves generalizing online convex optimization to a multi-agent setting and applying recent tools from nonstochastic control derived for a single agent. We empirically evaluate our method on a model of an overactuated aircraft. We show that the distributed method is robust to failure and to adversarial perturbations in the dynamics.

Udari Madhushani, Abhimanyu Dubey, Naomi Ehrich Leonard et al.

The cooperative bandit problem is increasingly becoming relevant due to its applications in large-scale decision-making. However, most research for this problem focuses exclusively on the setting with perfect communication, whereas in most real-world distributed settings, communication is often over stochastic networks, with arbitrary corruptions and delays. In this paper, we study cooperative bandit learning under three typical real-world communication scenarios, namely, (a) message-passing over stochastic time-varying networks, (b) instantaneous reward-sharing over a network with random delays, and (c) message-passing with adversarially corrupted rewards, including byzantine communication. For each of these environments, we propose decentralized algorithms that achieve competitive performance, along with near-optimal guarantees on the incurred group regret as well. Furthermore, in the setting with perfect communication, we present an improved delayed-update algorithm that outperforms the existing state-of-the-art on various network topologies. Finally, we present tight network-dependent minimax lower bounds on the group regret. Our proposed algorithms are straightforward to implement and obtain competitive empirical performance.

Justin Lidard, Udari Madhushani, Naomi Ehrich Leonard

A challenge in reinforcement learning (RL) is minimizing the cost of sampling associated with exploration. Distributed exploration reduces sampling complexity in multi-agent RL (MARL). We investigate the benefits to performance in MARL when exploration is fully decentralized. Specifically, we consider a class of online, episodic, tabular $Q$-learning problems under time-varying reward and transition dynamics, in which agents can communicate in a decentralized manner.We show that group performance, as measured by the bound on regret, can be significantly improved through communication when each agent uses a decentralized message-passing protocol, even when limited to sending information up to its $γ$-hop neighbors. We prove regret and sample complexity bounds that depend on the number of agents, communication network structure and $γ.$ We show that incorporating more agents and more information sharing into the group learning scheme speeds up convergence to the optimal policy. Numerical simulations illustrate our results and validate our theoretical claims.

Udari Madhushani, Naomi Leonard

In cooperative bandits, a framework that captures essential features of collective sequential decision making, agents can minimize group regret, and thereby improve performance, by leveraging shared information. However, sharing information can be costly, which motivates developing policies that minimize group regret while also reducing the number of messages communicated by agents. Existing cooperative bandit algorithms obtain optimal performance when agents share information with their neighbors at \textit{every time step}, i.e., full communication. This requires $Θ(T)$ number of messages, where $T$ is the time horizon of the decision making process. We propose \textit{ComEx}, a novel cost-effective communication protocol in which the group achieves the same order of performance as full communication while communicating only $O(\log T)$ number of messages. Our key step is developing a method to identify and only communicate the information crucial to achieving optimal performance. Further we propose novel algorithms for several benchmark cooperative bandit frameworks and show that our algorithms obtain \textit{state-of-the-art} performance while consistently incurring a significantly smaller communication cost than existing algorithms.

Udari Madhushani, Naomi Ehrich Leonard

We study the decentralized multi-agent multi-armed bandit problem for agents that communicate with probability over a network defined by a $d$-regular graph. Every edge in the graph has probabilistic weight $p$ to account for the ($1\!-\!p$) probability of a communication link failure. At each time step, each agent chooses an arm and receives a numerical reward associated with the chosen arm. After each choice, each agent observes the last obtained reward of each of its neighbors with probability $p$. We propose a new Upper Confidence Bound (UCB) based algorithm and analyze how agent-based strategies contribute to minimizing group regret in this probabilistic communication setting. We provide theoretical guarantees that our algorithm outperforms state-of-the-art algorithms. We illustrate our results and validate the theoretical claims using numerical simulations.

Udari Madhushani, Biswadip Dey, Naomi Ehrich Leonard et al.

Value function based reinforcement learning (RL) algorithms, for example, $Q$-learning, learn optimal policies from datasets of actions, rewards, and state transitions. However, when the underlying state transition dynamics are stochastic and evolve on a high-dimensional space, generating independent and identically distributed (IID) data samples for creating these datasets poses a significant challenge due to the intractability of the associated normalizing integral. In these scenarios, Hamiltonian Monte Carlo (HMC) sampling offers a computationally tractable way to generate data for training RL algorithms. In this paper, we introduce a framework, called \textit{Hamiltonian $Q$-Learning}, that demonstrates, both theoretically and empirically, that $Q$ values can be learned from a dataset generated by HMC samples of actions, rewards, and state transitions. Furthermore, to exploit the underlying low-rank structure of the $Q$ function, Hamiltonian $Q$-Learning uses a matrix completion algorithm for reconstructing the updated $Q$ function from $Q$ value updates over a much smaller subset of state-action pairs. Thus, by providing an efficient way to apply $Q$-learning in stochastic, high-dimensional settings, the proposed approach broadens the scope of RL algorithms for real-world applications.

Udari Madhushani, Naomi Leonard

We investigate the benefits of heterogeneity in multi-agent explore-exploit decision making where the goal of the agents is to maximize cumulative group reward. To do so we study a class of distributed stochastic bandit problems in which agents communicate over a multi-star network and make sequential choices among options in the same uncertain environment. Typically, in multi-agent bandit problems, agents use homogeneous decision-making strategies. However, group performance can be improved by incorporating heterogeneity into the choices agents make, especially when the network graph is irregular, i.e. when agents have different numbers of neighbors. We design and analyze new heterogeneous explore-exploit strategies, using the multi-star as the model irregular network graph. The key idea is to enable center agents to do more exploring than they would do using the homogeneous strategy, as a means of providing more useful data to the peripheral agents. In the case all agents broadcast their reward values and choices to their neighbors with the same probability, we provide theoretical guarantees that group performance improves under the proposed heterogeneous strategies as compared to under homogeneous strategies. We use numerical simulations to illustrate our results and to validate our theoretical bounds.

Udari Madhushani, Naomi Ehrich Leonard

We study cost-effective communication strategies that can be used to improve the performance of distributed learning systems in resource-constrained environments. For distributed learning in sequential decision making, we propose a new cost-effective partial communication protocol. We illustrate that with this protocol the group obtains the same order of performance that it obtains with full communication. Moreover, we prove that under the proposed partial communication protocol the communication cost is $O(\log T)$, where $T$ is the time horizon of the decision-making process. This improves significantly on protocols with full communication, which incur a communication cost that is $O(T)$. We validate our theoretical results using numerical simulations.

Udari Madhushani, Naomi Ehrich Leonard

We define and analyze a multi-agent multi-armed bandit problem in which decision-making agents can observe the choices and rewards of their neighbors under a linear observation cost. Neighbors are defined by a network graph that encodes the inherent observation constraints of the system. We define a cost associated with observations such that at every instance an agent makes an observation it receives a constant observation regret. We design a sampling algorithm and an observation protocol for each agent to maximize its own expected cumulative reward through minimizing expected cumulative sampling regret and expected cumulative observation regret. For our proposed protocol, we prove that total cumulative regret is logarithmically bounded. We verify the accuracy of analytical bounds using numerical simulations.

Udari Madhushani, Naomi Ehrich Leonard

We define and analyze a multi-agent multi-armed bandit problem in which decision-making agents can observe the choices and rewards of their neighbors. Neighbors are defined by a network graph with heterogeneous and stochastic interconnections. These interactions are determined by the sociability of each agent, which corresponds to the probability that the agent observes its neighbors. We design an algorithm for each agent to maximize its own expected cumulative reward and prove performance bounds that depend on the sociability of the agents and the network structure. We use the bounds to predict the rank ordering of agents according to their performance and verify the accuracy analytically and computationally.