We propose an extension of the reinforcement learning architecture that enables moral decision-making of reinforcement learning agents based on normative reasons. Central to this approach is a reason-based shield generator yielding a moral shield that binds the agent to actions that conform with recognized normative reasons so that our overall architecture restricts the agent to actions that are (internally) morally justified. In addition, we describe an algorithm that allows to iteratively improve the reason-based shield generator through case-based feedback from a moral judge.

As machine learning systems become more widely deployed, including in settings such as resume screening, hiring, lending, and recommendation systems, people have begun to respond to them strategically. Often, this takes the form of "gaming the system" or using an algorithmic system's rules and procedures to manipulate it and achieve desired outcomes. Examples include Uber drivers coordinating the times they log on and off the app to impact its surge pricing algorithm (Möhlmann and Zalmanson, 2017), and Twitter (Burrell et al., 2019) and Facebook (Eslami et al., 2016) users' decisions regarding how to interact with content given the platforms' curation algorithms. Game theoretical modeling and analysis have been used in recent years to formally analyze such strategic responses of humans to algorithms (e.g., Hardt et al. (2016); Milli et al. (2019); Liu et al. (2020); see also Related Work). However, these existing works assume standard models of decision making, where agents are fully rational when responding to algorithms; yet, humans exhibit different forms of cognitive biases in decision making (Kahnemann and Tversky, 1979). Motivated by this, we explore the impacts behavioral biases on agents' strategic responses to algorithms. We begin by proposing an extension of existing models of strategic classification to account for behavioral biases.

In this work, we study the problem of finding Pareto optimal policies in multi-agent reinforcement learning problems with cooperative reward structures. We show that any algorithm where each agent only optimizes their reward is subject to suboptimal convergence. Therefore, to achieve Pareto optimality, agents have to act altruistically by considering the rewards of others. This observation bridges the multi-objective optimization framework and multi-agent reinforcement learning together. We first propose a framework for applying the Multiple Gradient Descent algorithm (MGDA) for learning in multi-agent settings. We further show that standard MGDA is subjected to weak Pareto convergence, a problem that is often overlooked in other learning settings but is prevalent in multi-agent reinforcement learning. To mitigate this issue, we propose MGDA++, an improvement of the existing algorithm to handle the weakly optimal convergence of MGDA properly. Theoretically, we prove that MGDA++ converges to strong Pareto optimal solutions in convex, smooth bi-objective problems. We further demonstrate the superiority of our MGDA++ in cooperative settings in the Gridworld benchmark. The results highlight that our proposed method can converge efficiently and outperform the other methods in terms of the optimality of the convergent policies. The source code is available at \url{https://github.com/giangbang/Strong-Pareto-MARL}.

This paper focuses on decentralized stochastic bilevel optimization (DSBO) where agents only communicate with their neighbors. We propose Decentralized Stochastic Gradient Descent and Ascent with Gradient Tracking (DSGDA-GT), a novel algorithm that only requires first-order oracles that are much cheaper than second-order oracles widely adopted in existing works. We further provide a finite-time convergence analysis showing that for $n$ agents collaboratively solving the DSBO problem, the sample complexity of finding an $\epsilon$-stationary point in our algorithm is $\mathcal{O}(n^{-1}\epsilon^{-7})$, which matches the currently best-known results of the single-agent counterpart with linear speedup. The numerical experiments demonstrate both the communication and training efficiency of our algorithm.

The rapid development of large language and multimodal models has sparked significant interest in using proprietary models, such as GPT-4o, to develop autonomous agents capable of handling real-world scenarios like web navigation. Although recent open-source efforts have tried to equip agents with the ability to explore environments and continuously improve over time, they are building text-only agents in synthetic environments where the reward signals are clearly defined. Such agents struggle to generalize to realistic settings that require multimodal perception abilities and lack ground-truth signals. In this paper, we introduce an open-source framework designed to facilitate the development of multimodal web agent that can autonomously conduct real-world exploration and improve itself. We first train the base model with imitation learning to gain the basic abilities. We then let the agent explore the open web and collect feedback on its trajectories. After that, it further improves its policy by learning from well-performing trajectories judged by another general-purpose model. This exploration-feedback-optimization cycle can continue for several iterations. Experimental results show that our web agent successfully improves itself after each iteration, demonstrating strong performance across multiple test sets.

Shared control problems involve a robot learning to collaborate with a human. When learning a shared control policy, short communication between the agents can often significantly reduce running times and improve the system's accuracy. We extend the shared control problem to include the ability to directly query a cooperating agent. We consider two types of potential responses to a query, namely oracles: one that can provide the learner with the best action they should take, even when that action might be myopically wrong, and one with a bounded knowledge limited to its part of the system. Given this additional information channel, this work further presents three heuristics for choosing when to query: reinforcement learning-based, utility-based, and entropy-based. These heuristics aim to reduce a system's overall learning cost. Empirical results on two environments show the benefits of querying to learn a better control policy and the tradeoffs between the proposed heuristics.

Financial intelligence generation from vast data sources has typically relied on traditional methods of knowledge-graph construction or database engineering. Recently, fine-tuned financial domain-specific Large Language Models (LLMs), have emerged. While these advancements are promising, limitations such as high inference costs, hallucinations, and the complexity of concurrently analyzing high-dimensional financial data, emerge. This motivates our invention FISHNET (Financial Intelligence from Sub-querying, Harmonizing, Neural-Conditioning, Expert swarming, and Task planning), an agentic architecture that accomplishes highly complex analytical tasks for more than 98,000 regulatory filings that vary immensely in terms of semantics, data hierarchy, or format. FISHNET shows remarkable performance for financial insight generation (61.8% success rate over 5.0% Routing, 45.6% RAG R-Precision). We conduct rigorous ablations to empirically prove the success of FISHNET, each agent's importance, and the optimized performance of assembling all agents. Our modular architecture can be leveraged for a myriad of use-cases, enabling scalability, flexibility, and data integrity that are critical for financial tasks.

Designing effective two-sided matching mechanisms is a major problem in mechanism design, and the goodness of matching cannot always be formulated. The existing work addresses this issue by searching over a parameterized family of mechanisms with certain properties by learning to fit a human-crafted dataset containing examples of preference profiles and matching results. However, this approach does not consider a strategy-proof mechanism, implicitly assumes the number of agents to be a constant, and does not consider the public contextual information of the agents. In this paper, we propose a new parametric family of strategy-proof matching mechanisms by extending the serial dictatorship (SD). We develop a novel attention-based neural network called NeuralSD, which can learn a strategy-proof mechanism from a human-crafted dataset containing public contextual information. NeuralSD is constructed by tensor operations that make SD differentiable and learns a parameterized mechanism by estimating an order of SD from the contextual information. We conducted experiments to learn a strategy-proof matching from matching examples with different numbers of agents. We demonstrated that our method shows the superiority of learning with context-awareness over a baseline in terms of regression performance and other metrics.

Language agents have recently been used to simulate human behavior and user-item interactions for recommendation systems. However, current language agent simulations do not understand the relationships between users and items, leading to inaccurate user profiles and ineffective recommendations. In this work, we explore the utility of Knowledge Graphs (KGs), which contain extensive and reliable relationships between users and items, for recommendation. Our key insight is that the paths in a KG can capture complex relationships between users and items, eliciting the underlying reasons for user preferences and enriching user profiles. Leveraging this insight, we propose Knowledge Graph Enhanced Language Agents(KGLA), a framework that unifies language agents and KG for recommendation systems. In the simulated recommendation scenario, we position the user and item within the KG and integrate KG paths as natural language descriptions into the simulation. This allows language agents to interact with each other and discover sufficient rationale behind their interactions, making the simulation more accurate and aligned with real-world cases, thus improving recommendation performance. Our experimental results show that KGLA significantly improves recommendation performance (with a 33%-95% boost in NDCG@1 among three widely used benchmarks) compared to the previous best baseline method.

Enhancing the reasoning capabilities of large language models (LLMs) is crucial for enabling them to tackle complex, multi-step problems. Multi-agent frameworks have shown great potential in enhancing LLMs' reasoning capabilities. However, the lack of effective cooperation between LLM agents hinders their performance, especially for multi-step reasoning tasks. This paper proposes a novel cooperative multi-agent reasoning framework (CoPlanner) by separating reasoning steps and assigning distinct duties to different agents. CoPlanner consists of two LLM agents: a planning agent and a reasoning agent. The planning agent provides high-level strategic hints, while the reasoning agent follows these hints and infers answers. By training the planning agent's policy through the interactive reasoning process via Proximal Policy Optimization (PPO), the LLaMA-3-8B-based CoPlanner outperforms the previous best method by 9.94\% on LogiQA and 3.09\% on BBH. Our results demonstrate that the guidance from the planning agent and the effective cooperation between the agents contribute to the superior performance of CoPlanner in tackling multi-step reasoning problems.