Yue Jin

Mianchu Wang, Yue Jin, Giovanni Montana

Offline reinforcement learning (RL) aims to infer sequential decision policies using only offline datasets. This is a particularly difficult setup, especially when learning to achieve multiple different goals or outcomes under a given scenario with only sparse rewards. For offline learning of goal-conditioned policies via supervised learning, previous work has shown that an advantage weighted log-likelihood loss guarantees monotonic policy improvement. In this work we argue that, despite its benefits, this approach is still insufficient to fully address the distribution shift and multi-modality problems. The latter is particularly severe in long-horizon tasks where finding a unique and optimal policy that goes from a state to the desired goal is challenging as there may be multiple and potentially conflicting solutions. To tackle these challenges, we propose a complementary advantage-based weighting scheme that introduces an additional source of inductive bias: given a value-based partitioning of the state space, the contribution of actions expected to lead to target regions that are easier to reach, compared to the final goal, is further increased. Empirically, we demonstrate that the proposed approach, Dual-Advantage Weighted Offline Goal-conditioned RL (DAWOG), outperforms several competing offline algorithms in commonly used benchmarks. Analytically, we offer a guarantee that the learnt policy is never worse than the underlying behaviour policy.

Yunbo Qiu, Yuzhu Zhan, Yue Jin et al.

Flocking control is a significant problem in multi-agent systems such as multi-agent unmanned aerial vehicles and multi-agent autonomous underwater vehicles, which enhances the cooperativity and safety of agents. In contrast to traditional methods, multi-agent reinforcement learning (MARL) solves the problem of flocking control more flexibly. However, methods based on MARL suffer from sample inefficiency, since they require a huge number of experiences to be collected from interactions between agents and the environment. We propose a novel method Pretraining with Demonstrations for MARL (PwD-MARL), which can utilize non-expert demonstrations collected in advance with traditional methods to pretrain agents. During the process of pretraining, agents learn policies from demonstrations by MARL and behavior cloning simultaneously, and are prevented from overfitting demonstrations. By pretraining with non-expert demonstrations, PwD-MARL improves sample efficiency in the process of online MARL with a warm start. Experiments show that PwD-MARL improves sample efficiency and policy performance in the problem of flocking control, even with bad or few demonstrations.

Yunbo Qiu, Yue Jin, Jian Wang et al.

Flocking control is a challenging problem, where multiple agents, such as drones or vehicles, need to reach a target position while maintaining the flock and avoiding collisions with obstacles and collisions among agents in the environment. Multi-agent reinforcement learning has achieved promising performance in flocking control. However, methods based on traditional reinforcement learning require a considerable number of interactions between agents and the environment. This paper proposes a sub-optimal policy aided multi-agent reinforcement learning algorithm (SPA-MARL) to boost sample efficiency. SPA-MARL directly leverages a prior policy that can be manually designed or solved with a non-learning method to aid agents in learning, where the performance of the policy can be sub-optimal. SPA-MARL recognizes the difference in performance between the sub-optimal policy and itself, and then imitates the sub-optimal policy if the sub-optimal policy is better. We leverage SPA-MARL to solve the flocking control problem. A traditional control method based on artificial potential fields is used to generate a sub-optimal policy. Experiments demonstrate that SPA-MARL can speed up the training process and outperform both the MARL baseline and the used sub-optimal policy.

Yunbo Qiu, Yue Jin, Lebin Yu et al.

Multi-agent reinforcement learning (MARL) has achieved great progress in cooperative tasks in recent years. However, in the local reward scheme, where only local rewards for each agent are given without global rewards shared by all the agents, traditional MARL algorithms lack sufficient consideration of agents' mutual influence. In cooperative tasks, agents' mutual influence is especially important since agents are supposed to coordinate to achieve better performance. In this paper, we propose a novel algorithm Mutual-Help-based MARL (MH-MARL) to instruct agents to help each other in order to promote cooperation. MH-MARL utilizes an expected action module to generate expected other agents' actions for each particular agent. Then, the expected actions are delivered to other agents for selective imitation during training. Experimental results show that MH-MARL improves the performance of MARL both in success rate and cumulative reward.

Yue Jin, Shuangqing Wei, Jian Yuan et al.

Learning to coordinate is a daunting problem in multi-agent reinforcement learning (MARL). Previous works have explored it from many facets, including cognition between agents, credit assignment, communication, expert demonstration, etc. However, less attention were paid to agents' decision structure and the hierarchy of coordination. In this paper, we explore the spatiotemporal structure of agents' decisions and consider the hierarchy of coordination from the perspective of multilevel emergence dynamics, based on which a novel approach, Learning to Advise and Learning from Advice (LALA), is proposed to improve MARL. Specifically, by distinguishing the hierarchy of coordination, we propose to enhance decision coordination at meso level with an advisor and leverage a policy discriminator to advise agents' learning at micro level. The advisor learns to aggregate decision information in both spatial and temporal domains and generates coordinated decisions by employing a spatiotemporal dual graph convolutional neural network with a task-oriented objective function. Each agent learns from the advice via a policy generative adversarial learning method where a discriminator distinguishes between the policies of the agent and the advisor and boosts both of them based on its judgement. Experimental results indicate the advantage of LALA over baseline approaches in terms of both learning efficiency and coordination capability. Coordination mechanism is investigated from the perspective of multilevel emergence dynamics and mutual information point of view, which provides a novel perspective and method to analyze and improve MARL algorithms.

Hong Gao, Yiming Bao, Xuezhen Tu et al.

Video understanding requires not only visual recognition but also complex reasoning. While Vision-Language Models (VLMs) demonstrate impressive capabilities, they typically process videos largely in a single-pass manner with limited support for evidence revisit and iterative refinement. While recently emerging agent-based methods enable long-horizon reasoning, they either depend heavily on expensive proprietary models or require extensive agentic RL training. To overcome these limitations, we propose Agentic Video Intelligence (AVI), a flexible and training-free framework that can mirror human video comprehension through system-level design and optimization. AVI introduces three key innovations: (1) a human-inspired three-phase reasoning process (Retrieve-Perceive-Review) that ensures both sufficient global exploration and focused local analysis, (2) a structured video knowledge base organized through entity graphs, along with multi-granularity integrated tools, constituting the agent's interaction environment, and (3) an open-source model ensemble combining reasoning LLMs with lightweight base CV models and VLM, eliminating dependence on proprietary APIs or RL training. Experiments on LVBench, VideoMME-Long, LongVideoBench, and Charades-STA demonstrate that AVI achieves competitive performance while offering superior interpretability.

Yue Jin, Giovanni Montana

Offline multi-agent reinforcement learning (MARL) faces a critical challenge: the joint action space grows exponentially with the number of agents, making dataset coverage exponentially sparse and out-of-distribution (OOD) joint actions unavoidable. Partial Action Replacement (PAR) mitigates this by anchoring a subset of agents to dataset actions, but existing approach relies on enumerating multiple subset configurations at high computational cost and cannot adapt to varying states. We introduce PLCQL, a framework that formulates PAR subset selection as a contextual bandit problem and learns a state-dependent PAR policy using Proximal Policy Optimisation with an uncertainty-weighted reward. This adaptive policy dynamically determines how many agents to replace at each update step, balancing policy improvement against conservative value estimation. We prove a value-error bound showing that the estimation error scales linearly with the expected number of deviating agents. Compared with the previous PAR-based method SPaCQL, PLCQL reduces the number of per-iteration Q-function evaluations from n to 1, significantly improving computational efficiency. Empirically, PLCQL achieves the highest normalised scores on 66% of tasks across MPE, MaMuJoCo, and SMAC benchmarks, outperforming SPaCQL on 84% of tasks while substantially reducing computational cost.

Yue Jin, Giovanni Montana

Offline multi-agent reinforcement learning (MARL) is severely hampered by the challenge of evaluating out-of-distribution (OOD) joint actions. Our core finding is that when the behavior policy is factorized - a common scenario where agents act fully or partially independently during data collection - a strategy of partial action replacement (PAR) can significantly mitigate this challenge. PAR updates a single or part of agents' actions while the others remain fixed to the behavioral data, reducing distribution shift compared to full joint-action updates. Based on this insight, we develop Soft-Partial Conservative Q-Learning (SPaCQL), using PAR to mitigate OOD issue and dynamically weighting different PAR strategies based on the uncertainty of value estimation. We provide a rigorous theoretical foundation for this approach, proving that under factorized behavior policies, the induced distribution shift scales linearly with the number of deviating agents rather than exponentially with the joint-action space. This yields a provably tighter value error bound for this important class of offline MARL problems. Our theoretical results also indicate that SPaCQL adaptively addresses distribution shift using uncertainty-informed weights. Our empirical results demonstrate SPaCQL enables more effective policy learning, and manifest its remarkable superiority over baseline algorithms when the offline dataset exhibits the independence structure.

Mianchu Wang, Yue Jin, Giovanni Montana

Offline reinforcement learning (RL) seeks to learn optimal policies from static datasets without interacting with the environment. A common challenge is handling multi-modal action distributions, where multiple behaviours are represented in the data. Existing methods often assume unimodal behaviour policies, leading to suboptimal performance when this assumption is violated. We propose weighted imitation Learning on One Mode (LOM), a novel approach that focuses on learning from a single, promising mode of the behaviour policy. By using a Gaussian mixture model to identify modes and selecting the best mode based on expected returns, LOM avoids the pitfalls of averaging over conflicting actions. Theoretically, we show that LOM improves performance while maintaining simplicity in policy learning. Empirically, LOM outperforms existing methods on standard D4RL benchmarks and demonstrates its effectiveness in complex, multi-modal scenarios.

Sheng Tian, Xintan Zeng, Yifei Hu et al.

Graph-based patterns are extensively employed and favored by practitioners within industrial companies due to their capacity to represent the behavioral attributes and topological relationships among users, thereby offering enhanced interpretability in comparison to black-box models commonly utilized for classification and recognition tasks. For instance, within the scenario of transaction risk management, a graph pattern that is characteristic of a particular risk category can be readily employed to discern transactions fraught with risk, delineate networks of criminal activity, or investigate the methodologies employed by fraudsters. Nonetheless, graph data in industrial settings is often characterized by its massive scale, encompassing data sets with millions or even billions of nodes, making the manual extraction of graph patterns not only labor-intensive but also necessitating specialized knowledge in particular domains of risk. Moreover, existing methodologies for mining graph patterns encounter significant obstacles when tasked with analyzing large-scale attributed graphs. In this work, we introduce GraphRPM, an industry-purpose parallel and distributed risk pattern mining framework on large attributed graphs. The framework incorporates a novel edge-involved graph isomorphism network alongside optimized operations for parallel graph computation, which collectively contribute to a considerable reduction in computational complexity and resource expenditure. Moreover, the intelligent filtration of efficacious risky graph patterns is facilitated by the proposed evaluation metrics. Comprehensive experimental evaluations conducted on real-world datasets of varying sizes substantiate the capability of GraphRPM to adeptly address the challenges inherent in mining patterns from large-scale industrial attributed graphs, thereby underscoring its substantial value for industrial deployment.

Yue Jin, Yongchao Liu, Chuntao Hong

Graph-based computations are crucial in a wide range of applications, where graphs can scale to trillions of edges. To enable efficient training on such large graphs, mini-batch subgraph sampling is commonly used, which allows training without loading the entire graph into memory. However, existing solutions face significant trade-offs: online subgraph generation, as seen in frameworks like DGL and PyG, is limited to a single machine, resulting in severe performance bottlenecks, while offline precomputed subgraphs, as in GraphGen, improve sampling efficiency but introduce large storage overhead and high I/O costs during training. To address these challenges, we propose \textbf{GraphGen+}, an integrated framework that synchronizes distributed subgraph generation with in-memory graph learning, eliminating the need for external storage while significantly improving efficiency. GraphGen+ achieves a \textbf{27$\times$} speedup in subgraph generation compared to conventional SQL-like methods and a \textbf{1.3$\times$} speedup over GraphGen, supporting training on 1 million nodes per iteration and removing the overhead associated with precomputed subgraphs, making it a scalable and practical solution for industry-scale graph learning.

Yue Jin, Shuangqing Wei, Giovanni Montana

In human society, the conflict between self-interest and collective well-being often obstructs efforts to achieve shared welfare. Related concepts like the Tragedy of the Commons and Social Dilemmas frequently manifest in our daily lives. As artificial agents increasingly serve as autonomous proxies for humans, we propose a novel multi-agent reinforcement learning (MARL) method to address this issue - learning policies to maximise collective returns even when individual agents' interests conflict with the collective one. Unlike traditional cooperative MARL solutions that involve sharing rewards, values, and policies or designing intrinsic rewards to encourage agents to learn collectively optimal policies, we propose a novel MARL approach where agents exchange action suggestions. Our method reveals less private information compared to sharing rewards, values, or policies, while enabling effective cooperation without the need to design intrinsic rewards. Our algorithm is supported by our theoretical analysis that establishes a bound on the discrepancy between collective and individual objectives, demonstrating how sharing suggestions can align agents' behaviours with the collective objective. Experimental results demonstrate that our algorithm performs competitively with baselines that rely on value or policy sharing or intrinsic rewards.

Ting Zhu, Yue Jin, Jeremie Houssineau et al.

In decentralized multi-agent reinforcement learning, agents learning in isolation can lead to relative over-generalization (RO), where optimal joint actions are undervalued in favor of suboptimal ones. This hinders effective coordination in cooperative tasks, as agents tend to choose actions that are individually rational but collectively suboptimal. To address this issue, we introduce MaxMax Q-Learning (MMQ), which employs an iterative process of sampling and evaluating potential next states, selecting those with maximal Q-values for learning. This approach refines approximations of ideal state transitions, aligning more closely with the optimal joint policy of collaborating agents. We provide theoretical analysis supporting MMQ's potential and present empirical evaluations across various environments susceptible to RO. Our results demonstrate that MMQ frequently outperforms existing baselines, exhibiting enhanced convergence and sample efficiency.

Charles A. Hepburn, Yue Jin, Giovanni Montana

Traditional offline reinforcement learning (RL) methods predominantly operate in a batch-constrained setting. This confines the algorithms to a specific state-action distribution present in the dataset, reducing the effects of distributional shift but restricting the policy to seen actions. In this paper, we alleviate this limitation by introducing state-constrained offline RL, a novel framework that focuses solely on the dataset's state distribution. This approach allows the policy to take high-quality out-of-distribution actions that lead to in-distribution states, significantly enhancing learning potential. The proposed setting not only broadens the learning horizon but also improves the ability to combine different trajectories from the dataset effectively, a desirable property inherent in offline RL. Our research is underpinned by theoretical findings that pave the way for subsequent advancements in this area. Additionally, we introduce StaCQ, a deep learning algorithm that achieves state-of-the-art performance on the D4RL benchmark datasets and aligns with our theoretical propositions. StaCQ establishes a strong baseline for forthcoming explorations in this domain.

Yue Jin, Shuangqing Wei, Jian Yuan et al.

In multi-agent deep reinforcement learning, extracting sufficient and compact information of other agents is critical to attain efficient convergence and scalability of an algorithm. In canonical frameworks, distilling of such information is often done in an implicit and uninterpretable manner, or explicitly with cost functions not able to reflect the relationship between information compression and utility in representation. In this paper, we present Information-Bottleneck-based Other agents' behavior Representation learning for Multi-agent reinforcement learning (IBORM) to explicitly seek low-dimensional mapping encoder through which a compact and informative representation relevant to other agents' behaviors is established. IBORM leverages the information bottleneck principle to compress observation information, while retaining sufficient information relevant to other agents' behaviors used for cooperation decision. Empirical results have demonstrated that IBORM delivers the fastest convergence rate and the best performance of the learned policies, as compared with implicit behavior representation learning and explicit behavior representation learning without explicitly considering information compression and utility.

Yue Jin, Tianqing Zheng, Chao Gao et al.

Analyzing human affect is vital for human-computer interaction systems. Most methods are developed in restricted scenarios which are not practical for in-the-wild settings. The Affective Behavior Analysis in-the-wild (ABAW) 2021 Contest provides a benchmark for this in-the-wild problem. In this paper, we introduce a multi-modal and multi-task learning method by using both visual and audio information. We use both AU and expression annotations to train the model and apply a sequence model to further extract associations between video frames. We achieve an AU score of 0.712 and an expression score of 0.477 on the validation set. These results demonstrate the effectiveness of our approach in improving model performance.

Yue Jin, Yue Zhang, Tao Qin et al.

Off-policy evaluation (OPE) is to evaluate a target policy with data generated by other policies. Most previous OPE methods focus on precisely estimating the true performance of a policy. We observe that in many applications, (1) the end goal of OPE is to compare two or multiple candidate policies and choose a good one, which is a much simpler task than precisely evaluating their true performance; and (2) there are usually multiple policies that have been deployed to serve users in real-world systems and thus the true performance of these policies can be known. Inspired by the two observations, in this work, we study a new problem, supervised off-policy ranking (SOPR), which aims to rank a set of target policies based on supervised learning by leveraging off-policy data and policies with known performance. We propose a method to solve SOPR, which learns a policy scoring model by minimizing a ranking loss of the training policies rather than estimating the precise policy performance. The scoring model in our method, a hierarchical Transformer based model, maps a set of state-action pairs to a score, where the state of each pair comes from the off-policy data and the action is taken by a target policy on the state in an offline manner. Extensive experiments on public datasets show that our method outperforms baseline methods in terms of rank correlation, regret value, and stability. Our code is publicly available at GitHub.

Yongchao Liu, Yue Jin, Yong Chen et al.

Accelerating deep model training and inference is crucial in practice. Existing deep learning frameworks usually concentrate on optimizing training speed and pay fewer attentions to inference-specific optimizations. Actually, model inference differs from training in terms of computation, e.g. parameters are refreshed each gradient update step during training, but kept invariant during inference. These special characteristics of model inference open new opportunities for its optimization. In this paper, we propose a hardware-aware optimization framework, namely Woodpecker-DL (WPK), to accelerate inference by taking advantage of multiple joint optimizations from the perspectives of graph optimization, automated searches, domain-specific language (DSL) compiler techniques and system-level exploration. In WPK, we investigated two new automated search approaches based on genetic algorithm and reinforcement learning, respectively, to hunt the best operator code configurations targeting specific hardware. A customized DSL compiler is further attached to these search algorithms to generate efficient codes. To create an optimized inference plan, WPK systematically explores high-speed operator implementations from third-party libraries besides our automatically generated codes and singles out the best implementation per operator for use. Extensive experiments demonstrated that on a Tesla P100 GPU, we can achieve the maximum speedup of 5.40 over cuDNN and 1.63 over TVM on individual convolution operators, and run up to 1.18 times faster than TensorRT for end-to-end model inference.