Neural Information Processing Systems http://nips.cc/

One of the main challenges in reinforcement learning (RL) is generalisation. In typical deep RL methods this is achieved by approximating the optimal value function with a low-dimensional representation using a deep network. While this approach works well in many domains, in domains where the optimal value function cannot easily be reduced to a low-dimensional representation, learning can be very slow and unstable. This paper contributes towards tackling such challenging domains, by proposing a new method, called Hybrid Reward Architecture (HRA). HRA takes as input a decomposed reward function and learns a separate value function for each component reward function. Because each component typically only depends on a subset of all features, the corresponding value function can be approximated more easily by a low-dimensional representation, enabling more effective learning. We demonstrate HRA on a toy-problem and the Atari game Ms. Pac-Man, where HRA achieves above-human performance.

Long-Form Question Answering (LFQA) refers to generating in-depth, paragraph-level responses to open-ended questions. Although lots of LFQA methods are developed, evaluating LFQA effectively and efficiently remains challenging due to its high complexity and cost. Therefore, there is no standard benchmark for LFQA evaluation till now. To address this gap, we make the first attempt by proposing a well-constructed, reference-based benchmark named Chinese exAmination for LFQA Evaluation (CALF), aiming to rigorously assess the performance of automatic evaluation metrics for LFQA. The CALF benchmark is derived from Chinese examination questions that have been translated into English. It includes up to 1476 examples consisting of knowledge-intensive and nuanced responses. Our evaluation comprises three different settings to ana lyze the behavior of automatic metrics comprehensively. We conducted extensive experiments on 7 traditional evaluation metrics, 3 prompt-based metrics, and 3 trained evaluation metrics, and tested on agent systems for the LFQA evaluation. The results reveal that none of the current automatic evaluation metrics shows comparable performances with humans, indicating that they cannot capture dense information contained in long-form responses well. In addition, we provide a detailed analysis of the reasons why automatic evaluation metrics fail when evaluating LFQA, offering valuable insights to advance LFQA evaluation systems. Dataset and associated codes can be accessed at our GitHub repository.

The rapid proliferation of frontier model agents promises significant societal advances but also raises concerns about systemic risks arising from unsafe interactions. Collusion to the disadvantage of others has been identified as a central form of undesirable agent cooperation. The use of information hiding (steganography) in agent communications could render collusion practically undetectable. This underscores the need for evaluation frameworks to monitor and mitigate steganographic collusion capabilities. We address a crucial gap in the literature by demonstrating, for the first time, that robust steganographic collusion in LLMs can arise indirectly from optimization pressure. To investigate this problem we design two approaches -- a gradient-based reinforcement learning (GBRL) method and an in-context reinforcement learning (ICRL) method -- for reliably eliciting sophisticated LLM-generated linguistic text steganography. Importantly, we find that emergent steganographic collusion can be robust to both passive steganalytic oversight of model outputs and active mitigation through communication paraphrasing. We contribute a novel model evaluation framework and discuss limitations and future work. Our findings imply that effective risk mitigation from steganographic collusion post-deployment requires innovation in passive and active oversight techniques.

Analyzing individual human trajectory data helps our understanding of human mobility and finds many commercial and academic applications. There are two main approaches to accessing trajectory data for research: one involves using real-world datasets like GeoLife, while the other employs simulations to synthesize data. Real-world data provides insights from real human activities, but such data is generally sparse due to voluntary participation. Conversely, simulated data can be more comprehensive but may capture unrealistic human behavior. In this Data and Resource paper, we combine the benefit of both by leveraging the statistical features of real-world data and the comprehensiveness of simulated data. Specifically, we extract features from the real-world GeoLife dataset such as the average number of individual daily trips, average radius of gyration, and maximum and minimum trip distances. We calibrate the Pattern of Life Simulation, a realistic simulation of human mobility, to reproduce these features. Therefore, we use a genetic algorithm to calibrate the parameters of the simulation to mimic the GeoLife features. For this calibration, we simulated numerous random simulation settings, measured the similarity of generated trajectories to GeoLife, and iteratively (over many generations) combined parameter settings of trajectory datasets most similar to GeoLife. Using the calibrated simulation, we simulate large trajectory datasets that we call GeoLife+, where + denotes the Kleene Plus, indicating unlimited replication with at least one occurrence. We provide simulated GeoLife+ data with 182, 1k, and 5k over 5 years, 10k, and 50k over a year and 100k users over 6 months of simulation lifetime.

Agentic AIs $-$ AIs that are capable and permitted to undertake complex actions with little supervision $-$ mark a new frontier in AI capabilities and raise new questions about how to safely create and align such systems with users, developers, and society. Because agents' actions are influenced by their attitudes toward risk, one key aspect of alignment concerns the risk profiles of agentic AIs. Risk alignment will matter for user satisfaction and trust, but it will also have important ramifications for society more broadly, especially as agentic AIs become more autonomous and are allowed to control key aspects of our lives. AIs with reckless attitudes toward risk (either because they are calibrated to reckless human users or are poorly designed) may pose significant threats. They might also open 'responsibility gaps' in which there is no agent who can be held accountable for harmful actions. What risk attitudes should guide an agentic AI's decision-making? How might we design AI systems that are calibrated to the risk attitudes of their users? What guardrails, if any, should be placed on the range of permissible risk attitudes? What are the ethical considerations involved when designing systems that make risky decisions on behalf of others? We present three papers that bear on key normative and technical aspects of these questions.

Coding assistants are increasingly leveraged in game design, both generating code and making high-level plans. To what degree can these tools align with developer workflows, and what new modes of human-computer interaction can emerge from their use? We present DreamGarden, an AI system capable of assisting with the development of diverse game environments in Unreal Engine. At the core of our method is an LLM-driven planner, capable of breaking down a single, high-level prompt -- a dream, memory, or imagined scenario provided by a human user -- into a hierarchical action plan, which is then distributed across specialized submodules facilitating concrete implementation. This system is presented to the user as a garden of plans and actions, both growing independently and responding to user intervention via seed prompts, pruning, and feedback. Through a user study, we explore design implications of this system, charting courses for future work in semi-autonomous assistants and open-ended simulation design.

Open-ended short-answer questions (SAGs) have been widely recognized as a powerful tool for providing deeper insights into learners' responses in the context of learning analytics (LA). However, SAGs often present challenges in practice due to the high grading workload and concerns about inconsistent assessments. With recent advancements in natural language processing (NLP), automatic short-answer grading (ASAG) offers a promising solution to these challenges. Despite this, current ASAG algorithms are often limited in generalizability and tend to be tailored to specific questions. In this paper, we propose a unified multi-agent ASAG framework, GradeOpt, which leverages large language models (LLMs) as graders for SAGs. More importantly, GradeOpt incorporates two additional LLM-based agents - the reflector and the refiner - into the multi-agent system. This enables GradeOpt to automatically optimize the original grading guidelines by performing self-reflection on its errors. Through experiments on a challenging ASAG task, namely the grading of pedagogical content knowledge (PCK) and content knowledge (CK) questions, GradeOpt demonstrates superior performance in grading accuracy and behavior alignment with human graders compared to representative baselines. Finally, comprehensive ablation studies confirm the effectiveness of the individual components designed in GradeOpt.

In multi-agent learning, the predominant approach focuses on generalization, often neglecting the optimization of individual agents. This emphasis on generalization limits the ability of agents to utilize their unique strengths, resulting in inefficiencies. This paper introduces Comparative Advantage Maximization (CAM), a method designed to enhance individual agent specialization in multiagent systems. CAM employs a two-phase process, combining centralized population training with individual specialization through comparative advantage maximization. CAM achieved a 13.2% improvement in individual agent performance and a 14.9% increase in behavioral diversity compared to state-of-the-art systems. The success of CAM highlights the importance of individual agent specialization, suggesting new directions for multi-agent system development.

Offline reinforcement learning (RL) has garnered significant attention for its ability to learn effective policies from pre-collected datasets without the need for further environmental interactions. While promising results have been demonstrated in single-agent settings, offline multi-agent reinforcement learning (MARL) presents additional challenges due to the large joint state-action space and the complexity of multi-agent behaviors. A key issue in offline RL is the distributional shift, which arises when the target policy being optimized deviates from the behavior policy that generated the data. This problem is exacerbated in MARL due to the interdependence between agents' local policies and the expansive joint state-action space. Prior approaches have primarily addressed this challenge by incorporating regularization in the space of either Q-functions or policies. In this work, we introduce a regularizer in the space of stationary distributions to better handle distributional shift. Our algorithm, ComaDICE, offers a principled framework for offline cooperative MARL by incorporating stationary distribution regularization for the global learning policy, complemented by a carefully structured multi-agent value decomposition strategy to facilitate multi-agent training. Through extensive experiments on the multi-agent MuJoCo and StarCraft II benchmarks, we demonstrate that ComaDICE achieves superior performance compared to state-of-the-art offline MARL methods across nearly all tasks. Over the years, deep RL has achieved remarkable success in various decision-making tasks (Levine et al., 2016; Silver et al., 2017; Kalashnikov et al., 2018; Haydari & Yılmaz, 2020). However, a significant limitation of deep RL is its need for millions of interactions with the environment to gather experiences for policy improvement.