Shunyu Liu

Yu Wang, Tongya Zheng, Shunyu Liu et al.

Human mobility patterns have shown significant applications in policy-decision scenarios and economic behavior researches. The human mobility simulation task aims to generate human mobility trajectories given a small set of trajectory data, which have aroused much concern due to the scarcity and sparsity of human mobility data. Existing methods mostly rely on the static relationships of locations, while largely neglect the dynamic spatiotemporal effects of locations. On the one hand, spatiotemporal correspondences of visit distributions reveal the spatial proximity and the functionality similarity of locations. On the other hand, the varying durations in different locations hinder the iterative generation process of the mobility trajectory. Therefore, we propose a novel framework to model the dynamic spatiotemporal effects of locations, namely SpatioTemporal-Augmented gRaph neural networks (STAR). The STAR framework designs various spatiotemporal graphs to capture the spatiotemporal correspondences and builds a novel dwell branch to simulate the varying durations in locations, which is finally optimized in an adversarial manner. The comprehensive experiments over four real datasets for the human mobility simulation have verified the superiority of STAR to state-of-the-art methods. Our code is available at https://github.com/Star607/STAR-TKDE.

Shunyu Liu, Yihe Zhou, Jie Song et al.

Value Decomposition (VD) aims to deduce the contributions of agents for decentralized policies in the presence of only global rewards, and has recently emerged as a powerful credit assignment paradigm for tackling cooperative Multi-Agent Reinforcement Learning (MARL) problems. One of the main challenges in VD is to promote diverse behaviors among agents, while existing methods directly encourage the diversity of learned agent networks with various strategies. However, we argue that these dedicated designs for agent networks are still limited by the indistinguishable VD network, leading to homogeneous agent behaviors and thus downgrading the cooperation capability. In this paper, we propose a novel Contrastive Identity-Aware learning (CIA) method, explicitly boosting the credit-level distinguishability of the VD network to break the bottleneck of multi-agent diversity. Specifically, our approach leverages contrastive learning to maximize the mutual information between the temporal credits and identity representations of different agents, encouraging the full expressiveness of credit assignment and further the emergence of individualities. The algorithm implementation of the proposed CIA module is simple yet effective that can be readily incorporated into various VD architectures. Experiments on the SMAC benchmarks and across different VD backbones demonstrate that the proposed method yields results superior to the state-of-the-art counterparts. Our code is available at https://github.com/liushunyu/CIA.

Shunyu Liu, Jie Song, Yihe Zhou et al.

Deep cooperative multi-agent reinforcement learning has demonstrated its remarkable success over a wide spectrum of complex control tasks. However, recent advances in multi-agent learning mainly focus on value decomposition while leaving entity interactions still intertwined, which easily leads to over-fitting on noisy interactions between entities. In this work, we introduce a novel interactiOn Pattern disenTangling (OPT) method, to disentangle the entity interactions into interaction prototypes, each of which represents an underlying interaction pattern within a subgroup of the entities. OPT facilitates filtering the noisy interactions between irrelevant entities and thus significantly improves generalizability as well as interpretability. Specifically, OPT introduces a sparse disagreement mechanism to encourage sparsity and diversity among discovered interaction prototypes. Then the model selectively restructures these prototypes into a compact interaction pattern by an aggregator with learnable weights. To alleviate the training instability issue caused by partial observability, we propose to maximize the mutual information between the aggregation weights and the history behaviors of each agent. Experiments on single-task, multi-task and zero-shot benchmarks demonstrate that the proposed method yields results superior to the state-of-the-art counterparts. Our code is available at https://github.com/liushunyu/OPT.

Jiangtao Zhang, Shunyu Liu, Jie Song et al.

Weight Average (WA) is an active research topic due to its simplicity in ensembling deep networks and the effectiveness in promoting generalization. Existing weight average approaches, however, are often carried out along only one training trajectory in a post-hoc manner (i.e., the weights are averaged after the entire training process is finished), which significantly degrades the diversity between networks and thus impairs the effectiveness. In this paper, inspired by weight average, we propose Lookaround, a straightforward yet effective SGD-based optimizer leading to flatter minima with better generalization. Specifically, Lookaround iterates two steps during the whole training period: the around step and the average step. In each iteration, 1) the around step starts from a common point and trains multiple networks simultaneously, each on transformed data by a different data augmentation, and 2) the average step averages these trained networks to get the averaged network, which serves as the starting point for the next iteration. The around step improves the functionality diversity while the average step guarantees the weight locality of these networks during the whole training, which is essential for WA to work. We theoretically explain the superiority of Lookaround by convergence analysis, and make extensive experiments to evaluate Lookaround on popular benchmarks including CIFAR and ImageNet with both CNNs and ViTs, demonstrating clear superiority over state-of-the-arts. Our code is available at https://github.com/Ardcy/Lookaround.

Shunyu Liu, Yunpeng Qing, Shuqi Xu et al.

Inverse Reinforcement Learning (IRL) aims to reconstruct the reward function from expert demonstrations to facilitate policy learning, and has demonstrated its remarkable success in imitation learning. To promote expert-like behavior, existing IRL methods mainly focus on learning global reward functions to minimize the trajectory difference between the imitator and the expert. However, these global designs are still limited by the redundant noise and error propagation problems, leading to the unsuitable reward assignment and thus downgrading the agent capability in complex multi-stage tasks. In this paper, we propose a novel Curricular Subgoal-based Inverse Reinforcement Learning (CSIRL) framework, that explicitly disentangles one task with several local subgoals to guide agent imitation. Specifically, CSIRL firstly introduces decision uncertainty of the trained agent over expert trajectories to dynamically select subgoals, which directly determines the exploration boundary of different task stages. To further acquire local reward functions for each stage, we customize a meta-imitation objective based on these curricular subgoals to train an intrinsic reward generator. Experiments on the D4RL and autonomous driving benchmarks demonstrate that the proposed methods yields results superior to the state-of-the-art counterparts, as well as better interpretability. Our code is available at https://github.com/Plankson/CSIRL.

Yunpeng Qing, Shunyu Liu, Jie Song et al.

Reinforcement Learning (RL) is a popular machine learning paradigm where intelligent agents interact with the environment to fulfill a long-term goal. Driven by the resurgence of deep learning, Deep RL (DRL) has witnessed great success over a wide spectrum of complex control tasks. Despite the encouraging results achieved, the deep neural network-based backbone is widely deemed as a black box that impedes practitioners to trust and employ trained agents in realistic scenarios where high security and reliability are essential. To alleviate this issue, a large volume of literature devoted to shedding light on the inner workings of the intelligent agents has been proposed, by constructing intrinsic interpretability or post-hoc explainability. In this survey, we provide a comprehensive review of existing works on eXplainable RL (XRL) and introduce a new taxonomy where prior works are clearly categorized into model-explaining, reward-explaining, state-explaining, and task-explaining methods. We also review and highlight RL methods that conversely leverage human knowledge to promote learning efficiency and performance of agents while this kind of method is often ignored in XRL field. Some challenges and opportunities in XRL are discussed. This survey intends to provide a high-level summarization of XRL and to motivate future research on more effective XRL solutions. Corresponding open source codes are collected and categorized at https://github.com/Plankson/awesome-explainable-reinforcement-learning.

Shuang Luo, Yinchuan Li, Shunyu Liu et al.

Generative Flow Networks (GFlowNets) aim to generate diverse trajectories from a distribution in which the final states of the trajectories are proportional to the reward, serving as a powerful alternative to reinforcement learning for exploratory control tasks. However, the individual-flow matching constraint in GFlowNets limits their applications for multi-agent systems, especially continuous joint-control problems. In this paper, we propose a novel Multi-Agent generative Continuous Flow Networks (MACFN) method to enable multiple agents to perform cooperative exploration for various compositional continuous objects. Technically, MACFN trains decentralized individual-flow-based policies in a centralized global-flow-based matching fashion. During centralized training, MACFN introduces a continuous flow decomposition network to deduce the flow contributions of each agent in the presence of only global rewards. Then agents can deliver actions solely based on their assigned local flow in a decentralized way, forming a joint policy distribution proportional to the rewards. To guarantee the expressiveness of continuous flow decomposition, we theoretically derive a consistency condition on the decomposition network. Experimental results demonstrate that the proposed method yields results superior to the state-of-the-art counterparts and better exploration capability. Our code is available at https://github.com/isluoshuang/MACFN.

Yunpeng Qing, Yixiao Chi, Shuo Chen et al.

Recent advances in offline Reinforcement Learning (RL) have proven that effective policy learning can benefit from imposing conservative constraints on pre-collected datasets. However, such static datasets often exhibit distribution bias, resulting in limited generalizability. To address this limitation, a straightforward solution is data augmentation (DA), which leverages generative models to enrich data distribution. Despite the promising results, current DA techniques focus solely on reconstructing future trajectories from given states, while ignoring the exploration of history transitions that reach them. This single-direction paradigm inevitably hinders the discovery of diverse behavior patterns, especially those leading to critical states that may have yielded high-reward outcomes. In this work, we introduce Bidirectional Trajectory Diffusion (BiTrajDiff), a novel DA framework for offline RL that models both future and history trajectories from any intermediate states. Specifically, we decompose the trajectory generation task into two independent yet complementary diffusion processes: one generating forward trajectories to predict future dynamics, and the other generating backward trajectories to trace essential history transitions.BiTrajDiff can efficiently leverage critical states as anchors to expand into potentially valuable yet underexplored regions of the state space, thereby facilitating dataset diversity. Extensive experiments on the D4RL benchmark suite demonstrate that BiTrajDiff achieves superior performance compared to other advanced DA methods across various offline RL backbones.

Shunyu Liu, Kaixuan Chen, Na Yu et al.

Despite the promising results achieved, state-of-the-art interactive reinforcement learning schemes rely on passively receiving supervision signals from advisor experts, in the form of either continuous monitoring or pre-defined rules, which inevitably result in a cumbersome and expensive learning process. In this paper, we introduce a novel initiative advisor-in-the-loop actor-critic framework, termed as Ask-AC, that replaces the unilateral advisor-guidance mechanism with a bidirectional learner-initiative one, and thereby enables a customized and efficacious message exchange between learner and advisor. At the heart of Ask-AC are two complementary components, namely action requester and adaptive state selector, that can be readily incorporated into various discrete actor-critic architectures. The former component allows the agent to initiatively seek advisor intervention in the presence of uncertain states, while the latter identifies the unstable states potentially missed by the former especially when environment changes, and then learns to promote the ask action on such states. Experimental results on both stationary and non-stationary environments and across different actor-critic backbones demonstrate that the proposed framework significantly improves the learning efficiency of the agent, and achieves the performances on par with those obtained by continuous advisor monitoring.

Yuchen Ying, Weiqi Jiang, Tongya Zheng et al.

Knowledge graphs provide structured and reliable information for many real-world applications, motivating increasing interest in combining large language models (LLMs) with graph-based retrieval to improve factual grounding. Recent Graph-based Retrieval-Augmented Generation (GraphRAG) methods therefore introduce iterative interaction between LLMs and knowledge graphs to enhance reasoning capability. However, existing approaches typically depend on manually designed guidance and interact with knowledge graphs through a limited set of predefined tools, which substantially constrains graph exploration. To address these limitations, we propose GraphScout, a training-centric agentic graph reasoning framework equipped with more flexible graph exploration tools. GraphScout enables models to autonomously interact with knowledge graphs to synthesize structured training data which are then used to post-train LLMs, thereby internalizing agentic graph reasoning ability without laborious manual annotation or task curation. Extensive experiments across five knowledge-graph domains show that a small model (e.g., Qwen3-4B) augmented with GraphScout outperforms baseline methods built on leading LLMs (e.g., Qwen-Max) by an average of 16.7\% while requiring significantly fewer inference tokens. Moreover, GraphScout exhibits robust cross-domain transfer performance. Our code will be made publicly available~\footnote{https://github.com/Ying-Yuchen/_GraphScout_}.

Kongcheng Zhang, Qi Yao, Shunyu Liu et al.

Reinforcement Learning (RL) has shown promise for aligning Large Language Models (LLMs) to follow instructions with various constraints. Despite the encouraging results, RL improvement inevitably relies on sampling successful, high-quality responses; however, the initial model often struggles to generate responses that satisfy all constraints due to its limited capabilities, yielding sparse or indistinguishable rewards that impede learning. In this work, we propose Hindsight instruction Replay (HiR), a novel sample-efficient RL framework for complex instruction following tasks, which employs a select-then-rewrite strategy to replay failed attempts as successes based on the constraints that have been satisfied in hindsight. We perform RL on these replayed samples as well as the original ones, theoretically framing the objective as dual-preference learning at both the instruction- and response-level to enable efficient optimization using only a binary reward signal. Extensive experiments demonstrate that the proposed HiR yields promising results across different instruction following tasks, while requiring less computational budget. Our code and dataset is available at https://github.com/sastpg/HIR.

Wenda Li, Kaixuan Chen, Shunyu Liu et al.

Mini-batch Graph Transformer (MGT), as an emerging graph learning model, has demonstrated significant advantages in semi-supervised node prediction tasks with improved computational efficiency and enhanced model robustness. However, existing methods for processing local information either rely on sampling or simple aggregation, which respectively result in the loss and squashing of critical neighbor information.Moreover, the limited number of nodes in each mini-batch restricts the model's capacity to capture the global characteristic of the graph. In this paper, we propose LGMformer, a novel MGT model that employs a two-stage augmented interaction strategy, transitioning from local to global perspectives, to address the aforementioned bottlenecks.The local interaction augmentation (LIA) presents a neighbor-target interaction Transformer (NTIformer) to acquire an insightful understanding of the co-interaction patterns between neighbors and the target node, resulting in a locally effective token list that serves as input for the MGT. In contrast, global interaction augmentation (GIA) adopts a cross-attention mechanism to incorporate entire graph prototypes into the target node epresentation, thereby compensating for the global graph information to ensure a more comprehensive perception. To this end, LGMformer achieves the enhancement of node representations under the MGT paradigm.Experimental results related to node classification on the ten benchmark datasets demonstrate the effectiveness of the proposed method. Our code is available at https://github.com/l-wd/LGMformer.

Yibo Wang, Yongcheng Jing, Shunyu Liu et al.

Long-context reasoning has significantly empowered large language models (LLMs) to tackle complex tasks, yet it introduces severe efficiency bottlenecks due to the computational complexity. Existing efficient approaches often rely on complex additional training or external models for compression, which limits scalability and discards critical fine-grained information. In this paper, we propose VTC-R1, a new efficient reasoning paradigm that integrates vision-text compression into the reasoning process. Instead of processing lengthy textual traces, VTC-R1 renders intermediate reasoning segments into compact images, which are iteratively fed back into vision-language models as "optical memory." We construct a training dataset based on OpenR1-Math-220K achieving 3.4x token compression and fine-tune representative VLMs-Glyph and Qwen3-VL. Extensive experiments on benchmarks such as MATH500, AIME25, AMC23 and GPQA-D demonstrate that VTC-R1 consistently outperforms standard long-context reasoning. Furthermore, our approach significantly improves inference efficiency, achieving 2.7x speedup in end-to-end latency, highlighting its potential as a scalable solution for reasoning-intensive applications. Our code is available at https://github.com/w-yibo/VTC-R1.

Kaixuan Chen, Shunyu Liu, Na Yu et al.

The real-time transient stability assessment (TSA) plays a critical role in the secure operation of the power system. Although the classic numerical integration method, \textit{i.e.} time-domain simulation (TDS), has been widely used in industry practice, it is inevitably trapped in a high computational complexity due to the high latitude sophistication of the power system. In this work, a data-driven power system estimation method is proposed to quickly predict the stability of the power system before TDS reaches the end of simulating time windows, which can reduce the average simulation time of stability assessment without loss of accuracy. As the topology of the power system is in the form of graph structure, graph neural network based representation learning is naturally suitable for learning the status of the power system. Motivated by observing the distribution information of crucial active power and reactive power on the power system's bus nodes, we thus propose a distribution-aware learning~(DAL) module to explore an informative graph representation vector for describing the status of a power system. Then, TSA is re-defined as a binary classification task, and the stability of the system is determined directly from the resulting graph representation without numerical integration. Finally, we apply our method to the online TSA task. The case studies on the IEEE 39-bus system and Polish 2383-bus system demonstrate the effectiveness of our proposed method.

Yuwen Wang, Shunyu Liu, Kaixuan Chen et al.

Graph Lottery Ticket (GLT), a combination of core subgraph and sparse subnetwork, has been proposed to mitigate the computational cost of deep Graph Neural Networks (GNNs) on large input graphs while preserving original performance. However, the winning GLTs in exisiting studies are obtained by applying iterative magnitude-based pruning (IMP) without re-evaluating and re-considering the pruned information, which disregards the dynamic changes in the significance of edges/weights during graph/model structure pruning, and thus limits the appeal of the winning tickets. In this paper, we formulate a conjecture, i.e., existing overlooked valuable information in the pruned graph connections and model parameters which can be re-grouped into GLT to enhance the final performance. Specifically, we propose an adversarial complementary erasing (ACE) framework to explore the valuable information from the pruned components, thereby developing a more powerful GLT, referred to as the ACE-GLT. The main idea is to mine valuable information from pruned edges/weights after each round of IMP, and employ the ACE technique to refine the GLT processing. Finally, experimental results demonstrate that our ACE-GLT outperforms existing methods for searching GLT in diverse tasks. Our code will be made publicly available.

Huanjin Yao, Jiaxing Huang, Wenhao Wu et al.

In this work, we aim to develop an MLLM that understands and solves questions by learning to create each intermediate step of the reasoning involved till the final answer. To this end, we propose Collective Monte Carlo Tree Search (CoMCTS), a new learning-to-reason method for MLLMs, which introduces the concept of collective learning into ``tree search'' for effective and efficient reasoning-path searching and learning. The core idea of CoMCTS is to leverage collective knowledge from multiple models to collaboratively conjecture, search and identify effective reasoning paths toward correct answers via four iterative operations including Expansion, Simulation and Error Positioning, Backpropagation, and Selection. Using CoMCTS, we construct Mulberry-260k, a multimodal dataset with a tree of rich, explicit and well-defined reasoning nodes for each question. With Mulberry-260k, we perform collective SFT to train our model, Mulberry, a series of MLLMs with o1-like step-by-step Reasoning and Reflection capabilities. Extensive experiments demonstrate the superiority of our proposed methods on various benchmarks. Code will be available at https://github.com/HJYao00/Mulberry

Shunyu Liu, Yaoru Li, Kongcheng Zhang et al.

Recent studies have delved into constructing generalist agents for open-world environments like Minecraft. Despite the encouraging results, existing efforts mainly focus on solving basic programmatic tasks, e.g., material collection and tool-crafting following the Minecraft tech-tree, treating the ObtainDiamond task as the ultimate goal. This limitation stems from the narrowly defined set of actions available to agents, requiring them to learn effective long-horizon strategies from scratch. Consequently, discovering diverse gameplay opportunities in the open world becomes challenging. In this work, we introduce Odyssey, a new framework that empowers Large Language Model (LLM)-based agents with open-world skills to explore the vast Minecraft world. Odyssey comprises three key parts: (1) An interactive agent with an open-world skill library that consists of 40 primitive skills and 183 compositional skills. (2) A fine-tuned LLaMA-3 model trained on a large question-answering dataset with 390k+ instruction entries derived from the Minecraft Wiki. (3) A new agent capability benchmark includes the long-term planning task, the dynamic-immediate planning task, and the autonomous exploration task. Extensive experiments demonstrate that the proposed Odyssey framework can effectively evaluate different capabilities of LLM-based agents. All datasets, model weights, and code are publicly available to motivate future research on more advanced autonomous agent solutions.

Yuwen Wang, Shunyu Liu, Tongya Zheng et al.

Graph Neural Networks (GNNs) have emerged as a prominent framework for graph mining, leading to significant advances across various domains. Stemmed from the node-wise representations of GNNs, existing explanation studies have embraced the subgraph-specific viewpoint that attributes the decision results to the salient features and local structures of nodes. However, graph-level tasks necessitate long-range dependencies and global interactions for advanced GNNs, deviating significantly from subgraph-specific explanations. To bridge this gap, this paper proposes a novel intrinsically interpretable scheme for graph classification, termed as Global Interactive Pattern (GIP) learning, which introduces learnable global interactive patterns to explicitly interpret decisions. GIP first tackles the complexity of interpretation by clustering numerous nodes using a constrained graph clustering module. Then, it matches the coarsened global interactive instance with a batch of self-interpretable graph prototypes, thereby facilitating a transparent graph-level reasoning process. Extensive experiments conducted on both synthetic and real-world benchmarks demonstrate that the proposed GIP yields significantly superior interpretability and competitive performance to~the state-of-the-art counterparts. Our code will be made publicly available.

Yaoquan Wei, Shunyu Liu, Jie Song et al.

Action advising endeavors to leverage supplementary guidance from expert teachers to alleviate the issue of sampling inefficiency in Deep Reinforcement Learning (DRL). Previous agent-specific action advising methods are hindered by imperfections in the agent itself, while agent-agnostic approaches exhibit limited adaptability to the learning agent. In this study, we propose a novel framework called Agent-Aware trAining yet Agent-Agnostic Action Advising (A7) to strike a balance between the two. The underlying concept of A7 revolves around utilizing the similarity of state features as an indicator for soliciting advice. However, unlike prior methodologies, the measurement of state feature similarity is performed by neither the error-prone learning agent nor the agent-agnostic advisor. Instead, we employ a proxy model to extract state features that are both discriminative (adaptive to the agent) and generally applicable (robust to agent noise). Furthermore, we utilize behavior cloning to train a model for reusing advice and introduce an intrinsic reward for the advised samples to incentivize the utilization of expert guidance. Experiments are conducted on the GridWorld, LunarLander, and six prominent scenarios from Atari games. The results demonstrate that A7 significantly accelerates the learning process and surpasses existing methods (both agent-specific and agent-agnostic) by a substantial margin. Our code will be made publicly available.

Wenda Li, Kaixuan Chen, Shunyu Liu et al.

In this paper, we strive to develop an interpretable GNNs' inference paradigm, termed MSInterpreter, which can serve as a plug-and-play scheme readily applicable to various GNNs' baselines. Unlike the most existing explanation methods, MSInterpreter provides a Message-passing Selection scheme(MSScheme) to select the critical paths for GNNs' message aggregations, which aims at reaching the self-explaination instead of post-hoc explanations. In detail, the elaborate MSScheme is designed to calculate weight factors of message aggregation paths by considering the vanilla structure and node embedding components, where the structure base aims at weight factors among node-induced substructures; on the other hand, the node embedding base focuses on weight factors via node embeddings obtained by one-layer GNN.Finally, we demonstrate the effectiveness of our approach on graph classification benchmarks.

Yu Wang, Tongya Zheng, Yuxuan Liang et al.

Human trajectory data produced by daily mobile devices has proven its usefulness in various substantial fields such as urban planning and epidemic prevention. In terms of the individual privacy concern, human trajectory simulation has attracted increasing attention from researchers, targeting at offering numerous realistic mobility data for downstream tasks. Nevertheless, the prevalent issue of data scarcity undoubtedly degrades the reliability of existing deep learning models. In this paper, we are motivated to explore the intriguing problem of mobility transfer across cities, grasping the universal patterns of human trajectories to augment the powerful Transformer with external mobility data. There are two crucial challenges arising in the knowledge transfer across cities: 1) how to transfer the Transformer to adapt for domain heterogeneity; 2) how to calibrate the Transformer to adapt for subtly different long-tail frequency distributions of locations. To address these challenges, we have tailored a Cross-city mObiLity trAnsformer (COLA) with a dedicated model-agnostic transfer framework by effectively transferring cross-city knowledge for human trajectory simulation. Firstly, COLA divides the Transformer into the private modules for city-specific characteristics and the shared modules for city-universal mobility patterns. Secondly, COLA leverages a lightweight yet effective post-hoc adjustment strategy for trajectory simulation, without disturbing the complex bi-level optimization of model-agnostic knowledge transfer. Extensive experiments of COLA compared to state-of-the-art single-city baselines and our implemented cross-city baselines have demonstrated its superiority and effectiveness. The code is available at https://github.com/Star607/Cross-city-Mobility-Transformer.

Shunyu Liu, Wenkai Fang, Zetian Hu et al.

Large Language Models (LLMs) have demonstrated unprecedented generative capabilities, yet their alignment with human values remains critical for ensuring helpful and harmless deployments. While Reinforcement Learning from Human Feedback (RLHF) has emerged as a powerful paradigm for aligning LLMs with human preferences, its reliance on complex reward modeling introduces inherent trade-offs in computational efficiency and training stability. In this context, Direct Preference Optimization (DPO) has recently gained prominence as a streamlined alternative that directly optimizes LLMs using human preferences, thereby circumventing the need for explicit reward modeling. Owing to its theoretical elegance and computational efficiency, DPO has rapidly attracted substantial research efforts exploring its various implementations and applications. However, this field currently lacks systematic organization and comparative analysis. In this survey, we conduct a comprehensive overview of DPO and introduce a novel taxonomy, categorizing previous works into four key dimensions: data strategy, learning framework, constraint mechanism, and model property. We further present a rigorous empirical analysis of DPO variants across standardized benchmarks. Additionally, we discuss real-world applications, open challenges, and future directions for DPO. This work delivers both a conceptual framework for understanding DPO and practical guidance for practitioners, aiming to advance robust and generalizable alignment paradigms. All collected resources are available and will be continuously updated at https://github.com/liushunyu/awesome-direct-preference-optimization.

Wenkai Fang, Shunyu Liu, Yang Zhou et al.

Recent advances have demonstrated the effectiveness of Reinforcement Learning (RL) in improving the reasoning capabilities of Large Language Models (LLMs). However, existing works inevitably rely on high-quality instructions and verifiable rewards for effective training, both of which are often difficult to obtain in specialized domains. In this paper, we propose Self-play Reinforcement Learning(SeRL) to bootstrap LLM training with limited initial data. Specifically, SeRL comprises two complementary modules: self-instruction and self-rewarding. The former module generates additional instructions based on the available data at each training step, employing robust online filtering strategies to ensure instruction quality, diversity, and difficulty. The latter module introduces a simple yet effective majority-voting mechanism to estimate response rewards for additional instructions, eliminating the need for external annotations. Finally, SeRL performs conventional RL based on the generated data, facilitating iterative self-play learning. Extensive experiments on various reasoning benchmarks and across different LLM backbones demonstrate that the proposed SeRL yields results superior to its counterparts and achieves performance on par with those obtained by high-quality data with verifiable rewards. Our code is available at https://github.com/wantbook-book/SeRL.

Kongcheng Zhang, Qi Yao, Baisheng Lai et al.

In this work, we propose Reinforced Functional Token Tuning (RFTT), a novel reinforced fine-tuning framework that empowers Large Language Models (LLMs) with self-play learn-to-reason capabilities. Unlike prior prompt-driven reasoning efforts, RFTT embeds a rich set of learnable functional tokens (e.g., <analyze>, <verify>, <refine>) directly into the model vocabulary, enabling chain-of-thought construction with diverse human-like reasoning behaviors. Specifically, RFTT comprises two phases: (1) supervised fine-tuning performs prompt-driven tree search to obtain self-generated training data annotated with functional tokens, which warms up the model to learn these tokens for reasoning; and (2) online reinforcement learning further allows the model to explore different reasoning pathways through functional token sampling without relying on prompts, thereby facilitating effective self-improvement for functional reasoning. Extensive experiments demonstrate the superiority of the proposed RFTT on mathematical benchmarks, significantly boosting Qwen-2.5-7B-Instruct (70.6% to 79.8%) and LLaMA-3.1-8B-Instruct (32.2% to 60.2%) on the MATH dataset. Moreover, the performance of RFTT consistently improves with more search rollouts at inference time. Our code is available at https://github.com/sastpg/RFTT.

Zhenbang Xiao, Yu Wang, Shunyu Liu et al.

The burdensome training costs on large-scale graphs have aroused significant interest in graph condensation, which involves tuning Graph Neural Networks (GNNs) on a small condensed graph for use on the large-scale original graph. Existing methods primarily focus on aligning key metrics between the condensed and original graphs, such as gradients, output distribution and trajectories of GNNs, yielding satisfactory performance on downstream tasks. However, these complex metrics necessitate intricate external parameters and can potentially disrupt the optimization process of the condensation graph, making the condensation process highly demanding and unstable. Motivated by the recent success of simplified models across various domains, we propose a simplified approach to metric alignment in graph condensation, aiming to reduce unnecessary complexity inherited from intricate metrics. We introduce the Simple Graph Condensation (SimGC) framework, which aligns the condensed graph with the original graph from the input layer to the prediction layer, guided by a pre-trained Simple Graph Convolution (SGC) model on the original graph. Importantly, SimGC eliminates external parameters and exclusively retains the target condensed graph during the condensation process. This straightforward yet effective strategy achieves a significant speedup of up to 10 times compared to existing graph condensation methods while performing on par with state-of-the-art baselines. Comprehensive experiments conducted on seven benchmark datasets demonstrate the effectiveness of SimGC in prediction accuracy, condensation time, and generalization capability. Our code is available at https://github.com/BangHonor/SimGC.

Yunpeng Qing, Shunyu liu, Jingyuan Cong et al.

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

Hanyang Yuan, Ning Tang, Tongya Zheng et al.

Diversified recommendation has attracted increasing attention from both researchers and practitioners, which can effectively address the homogeneity of recommended items. Existing approaches predominantly aim to infer the diversity of user preferences from observed user feedback. Nonetheless, due to inherent data biases, the observed data may not fully reflect user interests, where underexplored preferences can be overwhelmed or remain unmanifested. Failing to capture these preferences can lead to suboptimal diversity in recommendations. To fill this gap, this work aims to study diversified recommendation from a data-bias perspective. Inspired by the outstanding performance of large language models (LLMs) in zero-shot inference leveraging world knowledge, we propose a novel approach that utilizes LLMs' expertise to uncover underexplored user preferences from observed behavior, ultimately providing diverse and relevant recommendations. To achieve this, we first introduce Tree of Preferences (ToP), an innovative structure constructed to model user preferences from coarse to fine. ToP enables LLMs to systematically reason over the user's rationale behind their behavior, thereby uncovering their underexplored preferences. To guide diversified recommendations using uncovered preferences, we adopt a data-centric approach, identifying candidate items that match user preferences and generating synthetic interactions that reflect underexplored preferences. These interactions are integrated to train a general recommender for diversification. Moreover, we scale up overall efficiency by dynamically selecting influential users during optimization. Extensive evaluations of both diversity and relevance show that our approach outperforms existing methods in most cases and achieves near-optimal performance in others, with reasonable inference latency.

Kongcheng Zhang, Qi Yao, Shunyu Liu et al.

Recent advances of Reinforcement Learning (RL) have highlighted its potential in complex reasoning tasks, yet effective training often relies on external supervision, which limits the broader applicability. In this work, we propose a novel self-rewarding reinforcement learning framework to enhance Large Language Model (LLM) reasoning by leveraging the consistency of intermediate reasoning states across different reasoning trajectories. Our key insight is that correct responses often exhibit consistent trajectory patterns in terms of model likelihood: their intermediate reasoning states tend to converge toward their own final answers (high consistency) with minimal deviation toward other candidates (low volatility). Inspired by this observation, we introduce CoVo, an intrinsic reward mechanism that integrates Consistency and Volatility via a robust vector-space aggregation strategy, complemented by a curiosity bonus to promote diverse exploration. CoVo enables LLMs to perform RL in a self-rewarding manner, offering a scalable pathway for learning to reason without external supervision. Extensive experiments on diverse reasoning benchmarks show that CoVo achieves performance comparable to or even surpassing supervised RL. Our code is available at https://github.com/sastpg/CoVo.

Yang Zhou, Sunzhu Li, Shunyu Liu et al.

Recent advances in Large Language Models (LLMs) have underscored the potential of Reinforcement Learning (RL) to facilitate the emergence of reasoning capabilities. Despite the encouraging results, a fundamental dilemma persists as RL improvement relies on learning from high-quality samples, yet the exploration for such samples remains bounded by the inherent limitations of LLMs. This, in effect, creates an undesirable cycle in which what cannot be explored cannot be learned. In this work, we propose Rubric-Scaffolded Reinforcement Learning (RuscaRL), a novel instructional scaffolding framework designed to break the exploration bottleneck for general LLM reasoning. Specifically, RuscaRL introduces checklist-style rubrics as (1) explicit scaffolding for exploration during rollout generation, where different rubrics are provided as external guidance within task instructions to steer diverse high-quality responses. This guidance is gradually decayed over time, encouraging the model to internalize the underlying reasoning patterns; (2) verifiable rewards for exploitation during model training, where we can obtain robust LLM-as-a-Judge scores using rubrics as references, enabling effective RL on general reasoning tasks. Extensive experiments demonstrate the superiority of the proposed RuscaRL across various benchmarks, effectively expanding reasoning boundaries under the Best-of-N evaluation. Notably, RuscaRL significantly boosts Qwen2.5-7B-Instruct from 23.6 to 50.3 on HealthBench-500, surpassing GPT-4.1. Furthermore, our fine-tuned variant on Qwen3-30B-A3B-Instruct achieves 61.1 on HealthBench-500, outperforming leading LLMs including OpenAI-o3. Our code is available at https://github.com/IANNXANG/RuscaRL.

Qinghong Guo, Yu Wang, Ji Cao et al.

Road network representation learning (RNRL) has attracted increasing attention from both researchers and practitioners as various spatiotemporal tasks are emerging. Recent advanced methods leverage Graph Neural Networks (GNNs) and contrastive learning to characterize the spatial structure of road segments in a self-supervised paradigm. However, spatial heterogeneity and temporal dynamics of road networks raise severe challenges to the neighborhood smoothing mechanism of self-supervised GNNs. To address these issues, we propose a $\textbf{D}$ual-branch $\textbf{S}$patial-$\textbf{T}$emporal self-supervised representation framework for enhanced road representations, termed as DST. On one hand, DST designs a mix-hop transition matrix for graph convolution to incorporate dynamic relations of roads from trajectories. Besides, DST contrasts road representations of the vanilla road network against that of the hypergraph in a spatial self-supervised way. The hypergraph is newly built based on three types of hyperedges to capture long-range relations. On the other hand, DST performs next token prediction as the temporal self-supervised task on the sequences of traffic dynamics based on a causal Transformer, which is further regularized by differentiating traffic modes of weekdays from those of weekends. Extensive experiments against state-of-the-art methods verify the superiority of our proposed framework. Moreover, the comprehensive spatiotemporal modeling facilitates DST to excel in zero-shot learning scenarios.

Wenhao Sun, Rong-Cheng Tu, Yifu Ding et al.

Video diffusion transformers have achieved remarkable progress in high-quality video generation, but remain computationally expensive due to the quadratic complexity of attention over high-dimensional video sequences. Recent acceleration methods enhance the efficiency by exploiting the local sparsity of attention scores; yet they often struggle with accelerating the long-range computation. To address this problem, we propose VORTA, an acceleration framework with two novel components: 1) a sparse attention mechanism that efficiently captures long-range dependencies, and 2) a routing strategy that adaptively replaces full 3D attention with specialized sparse attention variants. VORTA achieves an end-to-end speedup $1.76\times$ without loss of quality on VBench. Furthermore, it can seamlessly integrate with various other acceleration methods, such as model caching and step distillation, reaching up to speedup $14.41\times$ with negligible performance degradation. VORTA demonstrates its efficiency and enhances the practicality of video diffusion transformers in real-world settings. Codes and weights are available at https://github.com/wenhao728/VORTA.

Shunyu Liu, Jie Zhou, Qunxi Zhu et al.

Aspect-Based Sentiment Analysis (ABSA) stands as a crucial task in predicting the sentiment polarity associated with identified aspects within text. However, a notable challenge in ABSA lies in precisely determining the aspects' boundaries (start and end indices), especially for long ones, due to users' colloquial expressions. We propose DiffusionABSA, a novel diffusion model tailored for ABSA, which extracts the aspects progressively step by step. Particularly, DiffusionABSA gradually adds noise to the aspect terms in the training process, subsequently learning a denoising process that progressively restores these terms in a reverse manner. To estimate the boundaries, we design a denoising neural network enhanced by a syntax-aware temporal attention mechanism to chronologically capture the interplay between aspects and surrounding text. Empirical evaluations conducted on eight benchmark datasets underscore the compelling advantages offered by DiffusionABSA when compared against robust baseline models. Our code is publicly available at https://github.com/Qlb6x/DiffusionABSA.

Junjie Zhang, Guozheng Ma, Shunyu Liu et al.

Recent advances in Reinforcement Learning (RL) have underscored its potential for incentivizing reasoning capabilities of Large Language Models (LLMs). However, existing step-level efforts suffer from costly annotations that limit domain coverage, while scalar scores further impose an information bottleneck, offering insufficient semantic bandwidth to improve intermediate decisions. Alternative language-critique approaches, which rely on frozen or external critics, provide richer textual feedback but lack the scalability needed for sustained policy improvement. In this work, we propose language-driven stepwise trajectory redirection, termed as STRIDE, a novel training framework that shifts process supervision from scalar rewards to learnable stepwise language feedback. Specifically, we co-train a generator and a generative verifier using only outcome-based rewards, eliminating external annotations, while delivering sustained policy improvement through jointly aligned verifier training. The verifier's stepwise language critiques explicitly localize and explain failures, enabling the generator to redirect reasoning trajectories at intermediate steps toward alternative decisions. The trajectory redirection design guarantees harmless policy improvement, even under noisy or suboptimal verifier feedback. Experiments on diverse reasoning benchmarks show that STRIDE significantly outperforms state-of-the-art baselines, as well as achieving breakthroughs on zero-pass-rate problems where scalar methods yield no learning signal in our ablation studies, demonstrating the effectiveness of learnable stepwise language feedback for enhancing LLM reasoning.

Jingyi Zhang, Tianyi Lin, Huanjin Yao et al.

In this work, we aim to develop effective data synthesis techniques that autonomously synthesize multimodal training data for enhancing MLLMs in solving complex real-world tasks. To this end, we propose Collective Adversarial Data Synthesis (CADS), a novel and general approach to synthesize high-quality, diverse and challenging multimodal data for MLLMs. The core idea of CADS is to leverage collective intelligence to ensure high-quality and diverse generation, while exploring adversarial learning to synthesize challenging samples for effectively driving model improvement. Specifically, CADS operates with two cyclic phases, i.e., Collective Adversarial Data Generation (CAD-Generate) and Collective Adversarial Data Judgment (CAD-Judge). CAD-Generate leverages collective knowledge to jointly generate new and diverse multimodal data, while CAD-Judge collaboratively assesses the quality of synthesized data. In addition, CADS introduces an Adversarial Context Optimization mechanism to optimize the generation context to encourage challenging and high-value data generation. With CADS, we construct MMSynthetic-20K and train our model R1-SyntheticVL, which demonstrates superior performance on various benchmarks.

Huanjin Yao, Ruifei Zhang, Jiaxing Huang et al.

With the recent emergence of revolutionary autonomous agentic systems, research community is witnessing a significant shift from traditional static, passive, and domain-specific AI agents toward more dynamic, proactive, and generalizable agentic AI. Motivated by the growing interest in agentic AI and its potential trajectory toward AGI, we present a comprehensive survey on Agentic Multimodal Large Language Models (Agentic MLLMs). In this survey, we explore the emerging paradigm of agentic MLLMs, delineating their conceptual foundations and distinguishing characteristics from conventional MLLM-based agents. We establish a conceptual framework that organizes agentic MLLMs along three fundamental dimensions: (i) Agentic internal intelligence functions as the system's commander, enabling accurate long-horizon planning through reasoning, reflection, and memory; (ii) Agentic external tool invocation, whereby models proactively use various external tools to extend their problem-solving capabilities beyond their intrinsic knowledge; and (iii) Agentic environment interaction further situates models within virtual or physical environments, allowing them to take actions, adapt strategies, and sustain goal-directed behavior in dynamic real-world scenarios. To further accelerate research in this area for the community, we compile open-source training frameworks, training and evaluation datasets for developing agentic MLLMs. Finally, we review the downstream applications of agentic MLLMs and outline future research directions for this rapidly evolving field. To continuously track developments in this rapidly evolving field, we will also actively update a public repository at https://github.com/HJYao00/Awesome-Agentic-MLLMs.

Junjie Zhang, Rushuai Yang, Shunyu Liu et al.

In this work, we establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning under the token-level Markov decision process, revealing that large language models indeed learn an implicit $Q$-function for inference. Through this theoretical lens, we demonstrate that the widely used beam search method suffers from unacceptable over-optimism, where inference errors are inevitably amplified due to inflated $Q$-value estimations of suboptimal steps. To address this limitation, we propose Supervised Optimism Correction(SOC), which introduces a simple yet effective auxiliary loss for token-level $Q$-value estimations during supervised fine-tuning. Specifically, the auxiliary loss employs implicit value regularization to boost model confidence in expert-demonstrated responses, thereby suppressing over-optimism toward insufficiently supervised responses. Extensive experiments on mathematical reasoning benchmarks, including GSM8K, MATH, and GAOKAO, showcase the superiority of the proposed SOC with beam search across a series of open-source models.

Jingyi Zhang, Jiaxing Huang, Huanjin Yao et al.

Recent studies generally enhance MLLMs' reasoning capabilities via supervised fine-tuning on high-quality chain-of-thought reasoning data, which often leads models to merely imitate successful reasoning paths without understanding what the wrong reasoning paths are. In this work, we aim to enhance the MLLMs' reasoning ability beyond passively imitating positive reasoning paths. To this end, we design Step-wise Group Relative Policy Optimization (StepGRPO), a new online reinforcement learning framework that enables MLLMs to self-improve reasoning ability via simple, effective and dense step-wise rewarding. Specifically, StepGRPO introduces two novel rule-based reasoning rewards: Step-wise Reasoning Accuracy Reward (StepRAR) and Step-wise Reasoning Validity Reward (StepRVR). StepRAR rewards the reasoning paths that contain necessary intermediate reasoning steps via a soft key-step matching technique, while StepRAR rewards reasoning paths that follow a well-structured and logically consistent reasoning process through a reasoning completeness and logic evaluation strategy. With the proposed StepGRPO, we introduce R1-VL, a series of MLLMs with outstanding capabilities in step-by-step reasoning. Extensive experiments over 8 benchmarks demonstrate the superiority of our methods.

Di Wang, Shunyu Liu, Wentao Jiang et al.

Multimodal large language models (MLLMs) have undergone rapid development in advancing geospatial scene understanding. Recent studies have sought to enhance the reasoning capabilities of remote sensing MLLMs, typically through cold-start training with elaborately curated chain-of-thought (CoT) data. However, this approach not only incurs substantial annotation costs but also introduces human biases that may limit the diversity of model reasoning. To address these challenges, we propose GeoZero, a framework that enables MLLMs to perform geospatial reasoning without any predefined CoT supervision. Specifically, we construct two datasets, GeoZero-Instruct and GeoZero-Hard. GeoZero-Instruct allows the model to acquire preliminary geospatial knowledge through supervised fine-tuning, while GeoZero-Hard stimulates deep reasoning during the subsequent reinforcement learning stage. Furthermore, we introduce Answer-Anchored Group Relative Policy Optimization (A$^2$GRPO), where the reasoning process is regularized by the model's own answers, encouraging diverse yet accurate thinking. Extensive experiments on multiple remote sensing vision-language benchmarks demonstrate that GeoZero not only surpasses existing state-of-the-art methods but also fosters universal emergent reasoning capabilities across diverse geospatial tasks. Code, data, and models will be publicly available at https://github.com/MiliLab/GeoZero.

Wenda Li, Tongya Zheng, Shunyu Liu et al.

Heterogeneous graphs are widely present in real-world complex networks, where the diversity of node and relation types leads to complex and rich semantics. Efforts for modeling complex relation semantics in heterogeneous graphs are restricted by the limitations of predefined semantic dependencies and the scarcity of supervised signals. The advanced pre-training and fine-tuning paradigm leverages graph structure to provide rich self-supervised signals, but introduces semantic gaps between tasks. Large Language Models (LLMs) offer significant potential to address the semantic issues of relations and tasks in heterogeneous graphs through their strong reasoning capabilities in textual modality, but their incorporation into heterogeneous graphs is largely limited by computational complexity. Therefore, in this paper, we propose an Efficient LLM-Aware (ELLA) framework for heterogeneous graphs, addressing the above issues. To capture complex relation semantics, we propose an LLM-aware Relation Tokenizer that leverages LLM to encode multi-hop, multi-type relations. To reduce computational complexity, we further employ a Hop-level Relation Graph Transformer, which help reduces the complexity of LLM-aware relation reasoning from exponential to linear. To bridge semantic gaps between pre-training and fine-tuning tasks, we introduce the fine-grained task-aware textual Chain-of-Thought (CoT) prompts. Extensive experiments on four heterogeneous graphs show that our proposed ELLA outperforms state-of-the-art methods in the performance and efficiency. In particular, ELLA scales up to 13b-parameter LLMs and achieves up to a 4x speedup compared with existing LLM-based methods. Our code is publicly available at https://github.com/l-wd/ELLA.

Chaoyang Zhou, Shunyu Liu, Zengmao Wang et al.

Reward models are critical for improving large language models (LLMs), particularly in reinforcement learning from human feedback (RLHF) or inference-time verification. Current reward modeling typically relies on scores of overall responses to learn the outcome rewards for the responses. However, since the response-level scores are coarse-grained supervision signals, the reward model struggles to identify the specific components within a response trajectory that truly correlate with the scores, leading to poor generalization on unseen responses. In this paper, we propose to leverage generation probabilities to establish reward consistency between processes in the response trajectory, which allows the response-level supervisory signal to propagate across processes, thereby providing additional fine-grained signals for reward learning. Building on analysis under the Bayesian framework, we develop an intra-trajectory consistency regularization to enforce that adjacent processes with higher next-token generation probability maintain more consistent rewards. We apply the proposed regularization to the advanced outcome reward model, improving its performance on RewardBench. Besides, we show that the reward model trained with the proposed regularization induces better DPO-aligned policies and achieves better best-of-N (BON) inference-time verification results. Our code is provided in https://github.com/chaoyang101/ICRM.

Feiyang Xu, Shunyu Liu, Yunpeng Qing et al.

Active Voltage Control (AVC) on the Power Distribution Networks (PDNs) aims to stabilize the voltage levels to ensure efficient and reliable operation of power systems. With the increasing integration of distributed energy resources, recent efforts have explored employing multi-agent reinforcement learning (MARL) techniques to realize effective AVC. Existing methods mainly focus on the acquisition of short-term AVC strategies, i.e., only learning AVC within the short-term training trajectories of a singular diurnal cycle. However, due to the dynamic nature of load demands and renewable energy, the operation states of real-world PDNs may exhibit significant distribution shifts across varying timescales (e.g., daily and seasonal changes). This can render those short-term strategies suboptimal or even obsolete when performing continuous AVC over extended periods. In this paper, we propose a novel temporal prototype-aware learning method, abbreviated as TPA, to learn time-adaptive AVC under short-term training trajectories. At the heart of TPA are two complementary components, namely multi-scale dynamic encoder and temporal prototype-aware policy, that can be readily incorporated into various MARL methods. The former component integrates a stacked transformer network to learn underlying temporal dependencies at different timescales of the PDNs, while the latter implements a learnable prototype matching mechanism to construct a dedicated AVC policy that can dynamically adapt to the evolving operation states. Experimental results on the AVC benchmark with different PDN sizes demonstrate that the proposed TPA surpasses the state-of-the-art counterparts not only in terms of control performance but also by offering model transferability. Our code is available at https://github.com/Canyizl/TPA-for-AVC.

Zhenbang Xiao, Yu Wang, Shunyu Liu et al.

Graph condensation has emerged as an intriguing technique to save the expensive training costs of Graph Neural Networks (GNNs) by substituting a condensed small graph with the original graph. Despite the promising results achieved, previous methods usually employ an entangled paradigm of redundant parameters (nodes, edges, GNNs), which incurs complex joint optimization during condensation. This paradigm has considerably impeded the scalability of graph condensation, making it challenging to condense extremely large-scale graphs and generate high-fidelity condensed graphs. Therefore, we propose to disentangle the condensation process into a two-stage GNN-free paradigm, independently condensing nodes and generating edges while eliminating the need to optimize GNNs at the same time. The node condensation module avoids the complexity of GNNs by focusing on node feature alignment with anchors of the original graph, while the edge translation module constructs the edges of the condensed nodes by transferring the original structure knowledge with neighborhood anchors. This simple yet effective approach achieves at least 10 times faster than state-of-the-art methods with comparable accuracy on medium-scale graphs. Moreover, the proposed DisCo can successfully scale up to the Ogbn-papers100M graph containing over 100 million nodes with flexible reduction rates and improves performance on the second-largest Ogbn-products dataset by over 5%. Extensive downstream tasks and ablation study on five common datasets further demonstrate the effectiveness of the proposed DisCo framework. Our code is available at https://github.com/BangHonor/DisCo.

Jiaqi Han, Jingwen Ye, Shunyu Liu et al.

The success of large language models has garnered widespread attention for model merging techniques, especially training-free methods which combine model capabilities within the parameter space. However, two challenges remain: (1) uniform treatment of all parameters leads to performance degradation; (2) search-based algorithms are often inefficient. In this paper, we present an innovative framework termed Reinforced Model Merging (RMM), which encompasses an environment and agent tailored for merging tasks. These components interact to execute layer-wise merging actions, aiming to search the optimal merging architecture. Notably, RMM operates without any gradient computations on the original models, rendering it feasible for edge devices. Furthermore, by utilizing data subsets during the evaluation process, we addressed the bottleneck in the reward feedback phase, thereby accelerating RMM by up to 100 times. Extensive experiments demonstrate that RMM achieves state-of-the-art performance across various vision and NLP datasets and effectively overcomes the limitations of the existing baseline methods. Our code is available at https://github.com/WuDiHJQ/Reinforced-Model-Merging.

Kaixuan Chen, Shunyu Liu, Tongtian Zhu et al.

Graph Neural Networks (GNNs) have emerged as a powerful category of learning architecture for handling graph-structured data. However, existing GNNs typically ignore crucial structural characteristics in node-induced subgraphs, which thus limits their expressiveness for various downstream tasks. In this paper, we strive to strengthen the representative capabilities of GNNs by devising a dedicated plug-and-play normalization scheme, termed as SUbgraph-sPEcific FactoR Embedded Normalization (SuperNorm), that explicitly considers the intra-connection information within each node-induced subgraph. To this end, we embed the subgraph-specific factor at the beginning and the end of the standard BatchNorm, as well as incorporate graph instance-specific statistics for improved distinguishable capabilities. In the meantime, we provide theoretical analysis to support that, with the elaborated SuperNorm, an arbitrary GNN is at least as powerful as the 1-WL test in distinguishing non-isomorphism graphs. Furthermore, the proposed SuperNorm scheme is also demonstrated to alleviate the over-smoothing phenomenon. Experimental results related to predictions of graph, node, and link properties on the eight popular datasets demonstrate the effectiveness of the proposed method. The code is available at https://github.com/chenchkx/SuperNorm.

Yifu Ding, Wentao Jiang, Shunyu Liu et al.

Tree of Thoughts (ToT) enhances Large Language Model (LLM) reasoning by structuring problem-solving as a spanning tree. However, recent methods focus on search accuracy while overlooking computational efficiency. The challenges of accelerating the ToT lie in the frequent switching of reasoning focus, and the redundant exploration of suboptimal solutions. To alleviate this dilemma, we propose Dynamic Parallel Tree Search (DPTS), a novel parallelism framework that aims to dynamically optimize the reasoning path in inference. It includes the Parallelism Streamline in the generation phase to build up a flexible and adaptive parallelism with arbitrary paths by fine-grained cache management and alignment. Meanwhile, the Search and Transition Mechanism filters potential candidates to dynamically maintain the reasoning focus on more possible solutions and have less redundancy. Experiments on Qwen-2.5 and Llama-3 with Math500 and GSM8K datasets show that DPTS significantly improves efficiency by 2-4x on average while maintaining or even surpassing existing reasoning algorithms in accuracy, making ToT-based reasoning more scalable and computationally efficient.

Shunyu Liu, Wei Luo, Yanzhen Zhou et al.

Transmission interface power flow adjustment is a critical measure to ensure the security and economy operation of power systems. However, conventional model-based adjustment schemes are limited by the increasing variations and uncertainties occur in power systems, where the adjustment problems of different transmission interfaces are often treated as several independent tasks, ignoring their coupling relationship and even leading to conflict decisions. In this paper, we introduce a novel data-driven deep reinforcement learning (DRL) approach, to handle multiple power flow adjustment tasks jointly instead of learning each task from scratch. At the heart of the proposed method is a multi-task attribution map (MAM), which enables the DRL agent to explicitly attribute each transmission interface task to different power system nodes with task-adaptive attention weights. Based on this MAM, the agent can further provide effective strategies to solve the multi-task adjustment problem with a near-optimal operation cost. Simulation results on the IEEE 118-bus system, a realistic 300-bus system in China, and a very large European system with 9241 buses demonstrate that the proposed method significantly improves the performance compared with several baseline methods, and exhibits high interpretability with the learnable MAM.

Sunzhu Li, Jiale Zhao, Miteto Wei et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has driven substantial progress in reasoning-intensive domains like mathematics. However, optimizing open-ended generation remains challenging due to the lack of ground truth. While rubric-based evaluation offers a structured proxy for verification, existing methods suffer from scalability bottlenecks and coarse criteria, resulting in a supervision ceiling effect. To address this, we propose an automated Coarse-to-Fine Rubric Generation framework. By synergizing principle-guided synthesis, multi-model aggregation, and difficulty evolution, our approach produces comprehensive and highly discriminative criteria capable of capturing the subtle nuances. Based on this framework, we introduce RubricHub, a large-scale ($\sim$110k) and multi-domain dataset. We validate its utility through a two-stage post-training pipeline comprising Rubric-based Rejection Sampling Fine-Tuning (RuFT) and Reinforcement Learning (RuRL). Experimental results demonstrate that RubricHub unlocks significant performance gains: our post-trained Qwen3-14B achieves state-of-the-art (SOTA) results on HealthBench (69.3), surpassing proprietary frontier models such as GPT-5. The code and data will be released soon.

Kaixuan Chen, Wei Luo, Shunyu Liu et al.

In this paper, we present a novel transformer architecture tailored for learning robust power system state representations, which strives to optimize power dispatch for the power flow adjustment across different transmission sections. Specifically, our proposed approach, named Powerformer, develops a dedicated section-adaptive attention mechanism, separating itself from the self-attention used in conventional transformers. This mechanism effectively integrates power system states with transmission section information, which facilitates the development of robust state representations. Furthermore, by considering the graph topology of power system and the electrical attributes of bus nodes, we introduce two customized strategies to further enhance the expressiveness: graph neural network propagation and multi-factor attention mechanism. Extensive evaluations are conducted on three power system scenarios, including the IEEE 118-bus system, a realistic 300-bus system in China, and a large-scale European system with 9241 buses, where Powerformer demonstrates its superior performance over several baseline methods.

Kedi Chen, Jie Zhou, Qin Chen et al.

Information extraction (IE) aims to extract complex structured information from the text. Numerous datasets have been constructed for various IE tasks, leading to time-consuming and labor-intensive data annotations. Nevertheless, most prevailing methods focus on training task-specific models, while the common knowledge among different IE tasks is not explicitly modeled. Moreover, the same phrase may have inconsistent labels in different tasks, which poses a big challenge for knowledge transfer using a unified model. In this study, we propose a regularization-based transfer learning method for IE (TIE) via an instructed graph decoder. Specifically, we first construct an instruction pool for datasets from all well-known IE tasks, and then present an instructed graph decoder, which decodes various complex structures into a graph uniformly based on corresponding instructions. In this way, the common knowledge shared with existing datasets can be learned and transferred to a new dataset with new labels. Furthermore, to alleviate the label inconsistency problem among various IE tasks, we introduce a task-specific regularization strategy, which does not update the gradients of two tasks with 'opposite direction'. We conduct extensive experiments on 12 datasets spanning four IE tasks, and the results demonstrate the great advantages of our proposed method

Jie Yang, Yuwen Wang, Kaixuan Chen et al.

Interpretable Graph Neural Networks (GNNs) aim to reveal the underlying reasoning behind model predictions, attributing their decisions to specific subgraphs that are informative. However, existing subgraph-based interpretable methods suffer from an overemphasis on local structure, potentially overlooking long-range dependencies within the entire graphs. Although recent efforts that rely on graph coarsening have proven beneficial for global interpretability, they inevitably reduce the graphs to a fixed granularity. Such an inflexible way can only capture graph connectivity at a specific level, whereas real-world graph tasks often exhibit relationships at varying granularities (e.g., relevant interactions in proteins span from functional groups, to amino acids, and up to protein domains). In this paper, we introduce a novel Tree-like Interpretable Framework (TIF) for graph classification, where plain GNNs are transformed into hierarchical trees, with each level featuring coarsened graphs of different granularity as tree nodes. Specifically, TIF iteratively adopts a graph coarsening module to compress original graphs (i.e., root nodes of trees) into increasingly coarser ones (i.e., child nodes of trees), while preserving diversity among tree nodes within different branches through a dedicated graph perturbation module. Finally, we propose an adaptive routing module to identify the most informative root-to-leaf paths, providing not only the final prediction but also the multi-granular interpretability for the decision-making process. Extensive experiments on the graph classification benchmarks with both synthetic and real-world datasets demonstrate the superiority of TIF in interpretability, while also delivering a competitive prediction performance akin to the state-of-the-art counterparts.