Offline reinforcement learning endeavors to leverage offline datasets to craft effective agent policy without online interaction, which imposes proper conservative constraints with the support of behavior policies to tackle the out-of-distribution problem. However, existing works often suffer from the constraint conflict issue when offline datasets are collected from multiple behavior policies, i.e., different behavior policies may exhibit inconsistent actions with distinct returns across the state space. To remedy this issue, recent advantage-weighted methods prioritize samples with high advantage values for agent training while inevitably ignoring the diversity of behavior policy. In this paper, we introduce a novel Advantage-Aware Policy Optimization (A2PO) method to explicitly construct advantage-aware policy constraints for offline learning under mixed-quality datasets. Specifically, A2PO employs a conditional variational auto-encoder to disentangle the action distributions of intertwined behavior policies by modeling the advantage values of all training data as conditional variables. Then the agent can follow such disentangled action distribution constraints to optimize the advantage-aware policy towards high advantage values. Extensive experiments conducted on both the single-quality and mixed-quality datasets of the D4RL benchmark demonstrate that A2PO yields results superior to the counterparts. Our code is available at https://github.com/Plankson/A2PO.

Hallucination in large language models (LLMs) remains a critical barrier to their safe deployment. Existing tool-augmented hallucination detection methods require pre-defined fixed verification strategies, which are crucial to the quality and effectiveness of tool calls. Some methods directly employ powerful closed-source LLMs such as GPT-4 as detectors, which are effective but too costly. To mitigate the cost issue, some methods adopt the teacher-student architecture and finetune open-source small models as detectors via agent tuning. However, these methods are limited by fixed strategies. When faced with a dynamically changing execution environment, they may lack adaptability and inappropriately call tools, ultimately leading to detection failure. To address the problem of insufficient strategy adaptability, we propose the innovative ``Learning to Evaluate and Adaptively Plan''(LEAP) framework, which endows an efficient student model with the dynamic learning and proactive correction capabilities of the teacher model. Specifically, our method formulates the hallucination detection problem as a dynamic strategy learning problem. We first employ a teacher model to generate trajectories within the dynamic learning loop and dynamically adjust the strategy based on execution failures. We then distill this dynamic planning capability into an efficient student model via agent tuning. Finally, during strategy execution, the student model adopts a proactive correction mechanism, enabling it to propose, review, and optimize its own verification strategies before execution. We demonstrate through experiments on three challenging benchmarks that our LEAP-tuned model outperforms existing state-of-the-art methods.

Recent Large Reasoning Models (LRMs) have achieved remarkable performance in solving complex problems via supervised fine-tuning (SFT) and reinforcement learning (RL). Although existing RL algorithms significantly enhance model accuracy, they still suffer from excessively lengthy responses and overthinking issues, resulting in increased inference latency and computational consumption, especially for simple tasks that require minimal reasoning. To address this, we propose a novel RL framework, DEPO, to reduce inefficient reasoning for models. Our method mainly consists of three core components: (1) an innovative advantage decoupled algorithm to guide model reduction of inefficient tokens; (2) a difficulty-aware length penalty to lower the overall length of model responses; (3) an advantage clipping method to prevent bias in policy optimization. In our experiments, applied to DeepSeek-Distill-Qwen-7B and DeepSeek-Distill-Qwen-1.5B as base models, DEPO achieves a significant reduction in sequence length by 39% and reduces excessive reasoning paths in inefficient tokens, while outperforming the base model in overall accuracy.

We will add more details of the results in the final version.

Offline reinforcement learning endeavors to leverage offline datasets to craft effective agent policy without online interaction, which imposes proper conservative constraints with the support of behavior policies to tackle the out-of-distribution problem. However, existing works often suffer from the constraint conflict issue when offline datasets are collected from multiple behavior policies, i.e., different behavior policies may exhibit inconsistent actions with distinct returns across the state space. To remedy this issue, recent advantage-weighted methods prioritize samples with high advantage values for agent training while inevitably ignoring the diversity of behavior policy. In this paper, we introduce a novel Advantage-Aware Policy Optimization (A2PO) method to explicitly construct advantage-aware policy constraints for offline learning under mixed-quality datasets. Specifically, A2PO employs a conditional variational auto-encoder to disentangle the action distributions of intertwined behavior policies by modeling the advantage values of all training data as conditional variables.

Offline reinforcement learning endeavors to leverage offline datasets to craft effective agent policy without online interaction, which imposes proper conservative constraints with the support of behavior policies to tackle the out-of-distribution problem. However, existing works often suffer from the constraint conflict issue when offline datasets are collected from multiple behavior policies, i.e., different behavior policies may exhibit inconsistent actions with distinct returns across the state space. To remedy this issue, recent advantage-weighted methods prioritize samples with high advantage values for agent training while inevitably ignoring the diversity of behavior policy. In this paper, we introduce a novel Advantage-Aware Policy Optimization (A2PO) method to explicitly construct advantage-aware policy constraints for offline learning under mixed-quality datasets. Specifically, A2PO employs a conditional variational auto-encoder to disentangle the action distributions of intertwined behavior policies by modeling the advantage values of all training data as conditional variables. Then the agent can follow such disentangled action distribution constraints to optimize the advantage-aware policy towards high advantage values. Extensive experiments conducted on both the single-quality and mixed-quality datasets of the D4RL benchmark demonstrate that A2PO yields results superior to the counterparts. Our code will be made publicly available.

Recent works have applied the Proximal Policy Optimization (PPO) to the multi-agent cooperative tasks, such as Independent PPO (IPPO); and vanilla Multi-agent PPO (MAPPO) which has a centralized value function. However, previous literature shows that MAPPO may not perform as well as Independent PPO (IPPO) and the Fine-tuned QMIX on Starcraft Multi-Agent Challenge (SMAC). MAPPO-Feature-Pruned (MAPPO-FP) improves the performance of MAPPO by the carefully designed agent-specific features, which may be not friendly to algorithmic utility. By contrast, we find that MAPPO may face the problem of \textit{The Policies Overfitting in Multi-agent Cooperation(POMAC)}, as they learn policies by the sampled advantage values. Then POMAC may lead to updating the multi-agent policies in a suboptimal direction and prevent the agents from exploring better trajectories. In this paper, to mitigate the multi-agent policies overfitting, we propose a novel policy regularization method, which disturbs the advantage values via random Gaussian noise. The experimental results show that our method outperforms the Fine-tuned QMIX, MAPPO-FP, and achieves SOTA on SMAC without agent-specific features. We open-source the code at \url{https://github.com/hijkzzz/noisy-mappo}.

The problem of constrained reinforcement learning (CRL) holds significant importance as it provides a framework for addressing critical safety satisfaction concerns in the field of reinforcement learning (RL). However, with the introduction of constraint satisfaction, the current CRL methods necessitate the utilization of second-order optimization or primal-dual frameworks with additional Lagrangian multipliers, resulting in increased complexity and inefficiency during implementation. To address these issues, we propose a novel first-order feasible method named Constrained Proximal Policy Optimization (CPPO). By treating the CRL problem as a probabilistic inference problem, our approach integrates the Expectation-Maximization framework to solve it through two steps: 1) calculating the optimal policy distribution within the feasible region (E-step), and 2) conducting a first-order update to adjust the current policy towards the optimal policy obtained in the E-step (M-step). We establish the relationship between the probability ratios and KL divergence to convert the E-step into a convex optimization problem. Furthermore, we develop an iterative heuristic algorithm from a geometric perspective to solve this problem. Additionally, we introduce a conservative update mechanism to overcome the constraint violation issue that occurs in the existing feasible region method. Empirical evaluations conducted in complex and uncertain environments validate the effectiveness of our proposed method, as it performs at least as well as other baselines.