Multi-agent interactions between Large Language Model (LLM) agents have shown major improvements on diverse reasoning tasks. However, these involve long generations from multiple models across several rounds, making them expensive. Moreover, these multi-agent approaches fail to provide a final, single model for efficient inference. To address this, we introduce MAGDi, a new method for structured distillation of the reasoning interactions between multiple LLMs into smaller LMs. MAGDi teaches smaller models by representing multi-agent interactions as graphs, augmenting a base student model with a graph encoder, and distilling knowledge using three objective functions: next-token prediction, a contrastive loss between correct and incorrect reasoning, and a graph-based objective to model the interaction structure. Experiments on seven widely-used commonsense and math reasoning benchmarks show that MAGDi improves the reasoning capabilities of smaller models, outperforming several methods that distill from a single teacher and multiple teachers. Moreover, MAGDi also demonstrates an order of magnitude higher efficiency over its teachers. We conduct extensive analyses to show that MAGDi (1) enhances the generalizability to out-of-domain tasks, (2) scales positively with the size and strength of the base student model, and (3) obtains larger improvements (via our multi-teacher training) when applying self-consistency - an inference technique that relies on model diversity.

Foundation models (FMs) such as large language models have revolutionized the field of AI by showing remarkable performance in various tasks. However, they exhibit numerous limitations that prevent their broader adoption in many real-world systems, which often require a higher bar for trustworthiness and usability. Since FMs are trained using loss functions aimed at reconstructing the training corpus in a self-supervised manner, there is no guarantee that the model's output aligns with users' preferences for a specific task at hand. In this survey paper, we propose a conceptual framework that encapsulates different modes by which agents could interact with FMs and guide them suitably for a set of tasks, particularly through knowledge augmentation and reasoning. Our framework elucidates agent role categories such as updating the underlying FM, assisting with prompting the FM, and evaluating the FM output. We also categorize several state-of-the-art approaches into agent interaction protocols, highlighting the nature and extent of involvement of the various agent roles. The proposed framework provides guidance for future directions to further realize the power of FMs in practical AI systems.

In the realm of power systems, the increasing involvement of residential users in load forecasting applications has heightened concerns about data privacy. Specifically, the load data can inadvertently reveal the daily routines of residential users, thereby posing a risk to their property security. While federated learning (FL) has been employed to safeguard user privacy by enabling model training without the exchange of raw data, these FL models have shown vulnerabilities to emerging attack techniques, such as Deep Leakage from Gradients and poisoning attacks. To counteract these, we initially employ a Secure-Aggregation (SecAgg) algorithm that leverages multiparty computation cryptographic techniques to mitigate the risk of gradient leakage. However, the introduction of SecAgg necessitates the deployment of additional sub-center servers for executing the multiparty computation protocol, thereby escalating computational complexity and reducing system robustness, especially in scenarios where one or more sub-centers are unavailable. To address these challenges, we introduce a Markovian Switching-based distributed training framework, the convergence of which is substantiated through rigorous theoretical analysis. The Distributed Markovian Switching (DMS) topology shows strong robustness towards the poisoning attacks as well. Case studies employing real-world power system load data validate the efficacy of our proposed algorithm. It not only significantly minimizes communication complexity but also maintains accuracy levels comparable to traditional FL methods, thereby enhancing the scalability of our load forecasting algorithm.

We study how to use guidance to improve the throughput of lifelong Multi-Agent Path Finding (MAPF). Previous studies have demonstrated that while incorporating guidance, such as highways, can accelerate MAPF algorithms, this often results in a trade-off with solution quality. In addition, how to generate good guidance automatically remains largely unexplored, with current methods falling short of surpassing manually designed ones. In this work, we introduce the directed guidance graph as a versatile representation of guidance for lifelong MAPF, framing Guidance Graph Optimization (GGO) as the task of optimizing its edge weights. We present two GGO algorithms to automatically generate guidance for arbitrary lifelong MAPF algorithms and maps. The first method directly solves GGO by employing CMA-ES, a black-box optimization algorithm. The second method, PIU, optimizes an update model capable of generating guidance, demonstrating the ability to transfer optimized guidance graphs to larger maps with similar layouts. Empirically, we show that (1) our guidance graphs improve the throughput of three representative lifelong MAPF algorithms in four benchmark maps, and (2) our update model can generate guidance graphs for as large as $93 \times 91$ maps and as many as 3000 agents.

The rapid advancement and adoption of generative AI (GenAI) technologies like ChatGPT signify the dawn of "The Age of AI." (Gates, 2023; Kissinger, Schmidt, & Huttenlocher, 2021) These developments mark a significant leap in the capabilities and adoption of AI systems. However, for many people, the swift and disorienting integration of AI into daily life raises many issues (Cugurullo & Acheampong, 2023; Fietta, Zecchinato, Stasi, Polato, & Monaro, 2022; Qasem, 2023). Concerns include the potential impacts on employment, privacy, and inequality, along with broader societal implications like human rights, mental health, and the preservation of democratic norms (Future of Life Institute, 2023; Prabhakaran, Mitchell, Gebru, & Gabriel, 2022; Shahriari & Shahriari, 2017; Stray, 2020). This article argues for the importance of wellbeing as a key objective in AI and for human-centered design (HCD) as a key methodology. Based on this framing, it shares a set of key challenges that will face designers of AI for wellbeing, or Positive AI. The idea that AI should support wellbeing is not uncommon. In 2018, Zuckerberg (2018) (CEO of Meta, previously Facebook) publicly stated that wellbeing should be the goal of AI. Further, in an interview Jan Leike (Wiblin, n.d.) (head of the'Superalignment' research lab at OpenAI) said AI optimization should align to "flourishing."

In this article, a biophysically realistic model of a soft octopus arm with internal musculature is presented. The modeling is motivated by experimental observations of sensorimotor control where an arm localizes and reaches a target. Major contributions of this article are: (i) development of models to capture the mechanical properties of arm musculature, the electrical properties of the arm peripheral nervous system (PNS), and the coupling of PNS with muscular contractions; (ii) modeling the arm sensory system, including chemosensing and proprioception; and (iii) algorithms for sensorimotor control, which include a novel feedback neural motor control law for mimicking target-oriented arm reaching motions, and a novel consensus algorithm for solving sensing problems such as locating a food source from local chemical sensory information (exogenous) and arm deformation information (endogenous). Several analytical results, including rest-state characterization and stability properties of the proposed sensing and motor control algorithms, are provided. Numerical simulations demonstrate the efficacy of our approach. Qualitative comparisons against observed arm rest shapes and target-oriented reaching motions are also reported.

Alan Turing proposed in 1950 a framework called an imitation game to decide if a machine could think. Using mathematics developed largely after Turing -- category theory -- we analyze a broader class of universal imitation games (UIGs), which includes static, dynamic, and evolutionary games. In static games, the participants are in a steady state. In dynamic UIGs, "learner" participants are trying to imitate "teacher" participants over the long run. In evolutionary UIGs, the participants are competing against each other in an evolutionary game, and participants can go extinct and be replaced by others with higher fitness. We use the framework of category theory -- in particular, two influential results by Yoneda -- to characterize each type of imitation game. Universal properties in categories are defined by initial and final objects. We characterize dynamic UIGs where participants are learning by inductive inference as initial algebras over well-founded sets, and contrast them with participants learning by conductive inference over the final coalgebra of non-well-founded sets. We briefly discuss the extension of our categorical framework for UIGs to imitation games on quantum computers.

Due to strong capabilities in conducting fluent, multi-turn conversations with users, Large Language Models (LLMs) have the potential to further improve the performance of Conversational Recommender System (CRS). Unlike the aimless chit-chat that LLM excels at, CRS has a clear target. So it is imperative to control the dialogue flow in the LLM to successfully recommend appropriate items to the users. Furthermore, user feedback in CRS can assist the system in better modeling user preferences, which has been ignored by existing studies. However, simply prompting LLM to conduct conversational recommendation cannot address the above two key challenges. In this paper, we propose Multi-Agent Conversational Recommender System (MACRS) which contains two essential modules. First, we design a multi-agent act planning framework, which can control the dialogue flow based on four LLM-based agents. This cooperative multi-agent framework will generate various candidate responses based on different dialogue acts and then choose the most appropriate response as the system response, which can help MACRS plan suitable dialogue acts. Second, we propose a user feedback-aware reflection mechanism which leverages user feedback to reason errors made in previous turns to adjust the dialogue act planning, and higher-level user information from implicit semantics. We conduct extensive experiments based on user simulator to demonstrate the effectiveness of MACRS in recommendation and user preferences collection. Experimental results illustrate that MACRS demonstrates an improvement in user interaction experience compared to directly using LLMs.

This paper considers the problem of multi-agent reinforcement learning (multi-agent RL), where multiple agents aim to make decisions simultaneously in an unfamiliar environment to maximize their individual cumulative rewards. Multi-agent RL is used not only in large-scale strategy games like Go [Silver et al., 2017], Poker [Brown and Sandholm, 2019] and MOBA games [Ye et al., 2020], but also in various real-world applications such as autonomous driving[Shalev-Shwartz et al., 2016], household power management [Chung et al., 2020], and computer networking[Bhattacharyya et al., 2019]. The sheer amount of computation needed for self-play-based learning in these applications often demands the algorithm to run in a distributed fashion where the communication cost becomes a bottleneck. In such circumstances, instead of syncing up after each single trajectory, a more practical alternative is to assign a larger batch of work for each machine to perform independently and sync up only sporadically. The need for infrequent communication could be hard constraints in applications such as autonomous driving [Shalev-Shwartz et al., 2016]. Deploying new policies to vehicle firmware takes weeks, while new data are being collected in millions of cars every second. These constraints render standard multi-agent RL algorithms that require altering the policy after each new data point impractical. In the scenarios discussed above, the agent needs to minimize the number of policy deployments while learning a good policy using (nearly) the same number of samples as its fully-adaptive counterparts.

Remarkable performance of large language models (LLMs) in a variety of tasks brings forth many opportunities as well as challenges of utilizing them in production settings. Towards practical adoption of LLMs, multi-agent systems hold great promise to augment, integrate, and orchestrate LLMs in the larger context of enterprise platforms that use existing proprietary data and models to tackle complex real-world tasks. Despite the tremendous success of these systems, current approaches rely on narrow, single-focus objectives for optimization and evaluation, often overlooking potential constraints in real-world scenarios, including restricted budgets, resources and time. Furthermore, interpreting, analyzing, and debugging these systems requires different components to be evaluated in relation to one another. This demand is currently not feasible with existing methodologies. In this postion paper, we introduce the concept of reasoning capacity as a unifying criterion to enable integration of constraints during optimization and establish connections among different components within the system, which also enable a more holistic and comprehensive approach to evaluation. We present a formal definition of reasoning capacity and illustrate its utility in identifying limitations within each component of the system. We then argue how these limitations can be addressed with a self-reflective process wherein human-feedback is used to alleviate shortcomings in reasoning and enhance overall consistency of the system.