Wenhao Li

Zihan Wang, Xiang Xu, Hongyuan Zha et al. · uw

Healthcare mechanisms are inseparable from the strategic provider response they induce: existing healthcare AI benchmarks hold this response fixed and so cannot evaluate mechanisms by the equilibrium they produce. We recast hospital mechanism design as program synthesis for language models: typed, inspectable rule programs are executed and scored by Medi-Sim, a multi-agent simulator with five strategic provider channels (coding, selection, delay, effort, triage). An incentive sweep recovers classical health-economics findings as adjacent regimes -- up-coding and low-complexity-patient selection under profit pressure, and Goodhart-style drift where measured performance becomes anti-correlated with true outcomes -- and a single audit lever exposes pressure migration: closing the coding channel more than doubles low-complexity selection. LLM-guided evolutionary code search over the same rule-program space then synthesizes an inspectable mixed-objective program that eliminates up-coding, halves rejection, and retains most of the profit-oriented baseline's funds.

Shuhui Zhu, Yue Lin, Shriya Kaistha et al.

Indirect reciprocity, which means helping those who have helped others, is difficult to sustain among decentralized, self-interested LLM agents without reliable reputation systems. We address this challenge with the Agentic Linguistic Gossip Network (ALIGN), an automated framework that enables decentralized agents to form reputations, evaluate trustworthiness, and coordinate social norms by strategically sharing open-ended gossip with hierarchical tones. We demonstrate that ALIGN consistently improves indirect reciprocity and resists malicious entrants by identifying and ostracizing defectors. Notably, we find that stronger reasoning capabilities in LLMs lead to more incentive-aligned cooperation, whereas chat models often over-cooperate even when strategically suboptimal. These results suggest that leveraging LLM reasoning through decentralized gossip is a promising path for maintaining social welfare in agentic ecosystems. Our code is available at https://github.com/shuhui-zhu/ALIGN.

Jinyi Hu, Xiaoyuan Yi, Wenhao Li et al. · tsinghua

The past several years have witnessed Variational Auto-Encoder's superiority in various text generation tasks. However, due to the sequential nature of the text, auto-regressive decoders tend to ignore latent variables and then reduce to simple language models, known as the KL vanishing problem, which would further deteriorate when VAE is combined with Transformer-based structures. To ameliorate this problem, we propose DELLA, a novel variational Transformer framework. DELLA learns a series of layer-wise latent variables with each inferred from those of lower layers and tightly coupled with the hidden states by low-rank tensor product. In this way, DELLA forces these posterior latent variables to be fused deeply with the whole computation path and hence incorporate more information. We theoretically demonstrate that our method can be regarded as entangling latent variables to avoid posterior information decrease through layers, enabling DELLA to get higher non-zero KL values even without any annealing or thresholding tricks. Experiments on four unconditional and three conditional generation tasks show that DELLA could better alleviate KL vanishing and improve both quality and diversity compared to several strong baselines.

Wenhao Li, Mengyuan Liu, Hong Liu et al.

Modern multi-layer perceptron (MLP) models have shown competitive results in learning visual representations without self-attention. However, existing MLP models are not good at capturing local details and lack prior knowledge of human body configurations, which limits their modeling power for skeletal representation learning. To address these issues, we propose a simple yet effective graph-reinforced MLP-Like architecture, named GraphMLP, that combines MLPs and graph convolutional networks (GCNs) in a global-local-graphical unified architecture for 3D human pose estimation. GraphMLP incorporates the graph structure of human bodies into an MLP model to meet the domain-specific demand of the 3D human pose, while allowing for both local and global spatial interactions. Furthermore, we propose to flexibly and efficiently extend the GraphMLP to the video domain and show that complex temporal dynamics can be effectively modeled in a simple way with negligible computational cost gains in the sequence length. To the best of our knowledge, this is the first MLP-Like architecture for 3D human pose estimation in a single frame and a video sequence. Extensive experiments show that the proposed GraphMLP achieves state-of-the-art performance on two datasets, i.e., Human3.6M and MPI-INF-3DHP. Code and models are available at https://github.com/Vegetebird/GraphMLP.

Ruiyi Wang, Wenhao Li, Xiaohong Liu et al.

Existing methods have achieved remarkable performance in image dehazing, particularly on synthetic datasets. However, they often struggle with real-world hazy images due to domain shift, limiting their practical applicability. This paper introduces HazeCLIP, a language-guided adaptation framework designed to enhance the real-world performance of pre-trained dehazing networks. Inspired by the Contrastive Language-Image Pre-training (CLIP) model's ability to distinguish between hazy and clean images, we leverage it to evaluate dehazing results. Combined with a region-specific dehazing technique and tailored prompt sets, the CLIP model accurately identifies hazy areas, providing a high-quality, human-like prior that guides the fine-tuning process of pre-trained networks. Extensive experiments demonstrate that HazeCLIP achieves state-of-the-art performance in real-word image dehazing, evaluated through both visual quality and image quality assessment metrics. Codes are available at https://github.com/Troivyn/HazeCLIP.

Wenhao Li, Mengyuan Liu, Hong Liu et al.

Transformers have been successfully applied in the field of video-based 3D human pose estimation. However, the high computational costs of these video pose transformers (VPTs) make them impractical on resource-constrained devices. In this paper, we present a plug-and-play pruning-and-recovering framework, called Hourglass Tokenizer (HoT), for efficient transformer-based 3D human pose estimation from videos. Our HoT begins with pruning pose tokens of redundant frames and ends with recovering full-length tokens, resulting in a few pose tokens in the intermediate transformer blocks and thus improving the model efficiency. To effectively achieve this, we propose a token pruning cluster (TPC) that dynamically selects a few representative tokens with high semantic diversity while eliminating the redundancy of video frames. In addition, we develop a token recovering attention (TRA) to restore the detailed spatio-temporal information based on the selected tokens, thereby expanding the network output to the original full-length temporal resolution for fast inference. Extensive experiments on two benchmark datasets (i.e., Human3.6M and MPI-INF-3DHP) demonstrate that our method can achieve both high efficiency and estimation accuracy compared to the original VPT models. For instance, applying to MotionBERT and MixSTE on Human3.6M, our HoT can save nearly 50% FLOPs without sacrificing accuracy and nearly 40% FLOPs with only 0.2% accuracy drop, respectively. Code and models are available at https://github.com/NationalGAILab/HoT.

Yingxuan You, Hong Liu, Ti Wang et al.

Despite significant progress in single image-based 3D human mesh recovery, accurately and smoothly recovering 3D human motion from a video remains challenging. Existing video-based methods generally recover human mesh by estimating the complex pose and shape parameters from coupled image features, whose high complexity and low representation ability often result in inconsistent pose motion and limited shape patterns. To alleviate this issue, we introduce 3D pose as the intermediary and propose a Pose and Mesh Co-Evolution network (PMCE) that decouples this task into two parts: 1) video-based 3D human pose estimation and 2) mesh vertices regression from the estimated 3D pose and temporal image feature. Specifically, we propose a two-stream encoder that estimates mid-frame 3D pose and extracts a temporal image feature from the input image sequence. In addition, we design a co-evolution decoder that performs pose and mesh interactions with the image-guided Adaptive Layer Normalization (AdaLN) to make pose and mesh fit the human body shape. Extensive experiments demonstrate that the proposed PMCE outperforms previous state-of-the-art methods in terms of both per-frame accuracy and temporal consistency on three benchmark datasets: 3DPW, Human3.6M, and MPI-INF-3DHP. Our code is available at https://github.com/kasvii/PMCE.

Jialun Cai, Hong Liu, Runwei Ding et al.

3D human pose estimation errors would propagate along the human body topology and accumulate at the end joints of limbs. Inspired by the backtracking mechanism in automatic control systems, we design an Intra-Part Constraint module that utilizes the parent nodes as the reference to build topological constraints for end joints at the part level. Further considering the hierarchy of the human topology, joint-level and body-level dependencies are captured via graph convolutional networks and self-attentions, respectively. Based on these designs, we propose a novel Human Topology aware Network (HTNet), which adopts a channel-split progressive strategy to sequentially learn the structural priors of the human topology from multiple semantic levels: joint, part, and body. Extensive experiments show that the proposed method improves the estimation accuracy by 18.7% on the end joints of limbs and achieves state-of-the-art results on Human3.6M and MPI-INF-3DHP datasets. Code is available at https://github.com/vefalun/HTNet.

Wenhao Li, Guangyang Wu, Wenyi Wang et al.

Low-Light Video Enhancement (LLVE) has received considerable attention in recent years. One of the critical requirements of LLVE is inter-frame brightness consistency, which is essential for maintaining the temporal coherence of the enhanced video. However, most existing single-image-based methods fail to address this issue, resulting in flickering effect that degrades the overall quality after enhancement. Moreover, 3D Convolution Neural Network (CNN)-based methods, which are designed for video to maintain inter-frame consistency, are computationally expensive, making them impractical for real-time applications. To address these issues, we propose an efficient pipeline named FastLLVE that leverages the Look-Up-Table (LUT) technique to maintain inter-frame brightness consistency effectively. Specifically, we design a learnable Intensity-Aware LUT (IA-LUT) module for adaptive enhancement, which addresses the low-dynamic problem in low-light scenarios. This enables FastLLVE to perform low-latency and low-complexity enhancement operations while maintaining high-quality results. Experimental results on benchmark datasets demonstrate that our method achieves the State-Of-The-Art (SOTA) performance in terms of both image quality and inter-frame brightness consistency. More importantly, our FastLLVE can process 1,080p videos at $\mathit{50+}$ Frames Per Second (FPS), which is $\mathit{2 \times}$ faster than SOTA CNN-based methods in inference time, making it a promising solution for real-time applications. The code is available at https://github.com/Wenhao-Li-777/FastLLVE.

Wenhao Li, Xiaoyuan Yi, Jinyi Hu et al. · tsinghua

Recently, powerful Transformer architectures have proven superior in generating high-quality sentences. Nevertheless, these models tend to produce dull high-frequency phrases, severely hurting the diversity and novelty of generated text. In this work, we dig into the intrinsic mechanism of this problem and found that sparser attention values in Transformer could improve diversity. To understand such a phenomenon, we first conduct both empirical and theoretical analysis and then attribute it to representation degeneration caused by the attentive mixture of the hidden states during training. We term this process the Trap of Mediocrity. To escape from such a trap, we introduce a novel attention regularization loss to control the sharpness of the attention distribution, which is transparent to model structures and can be easily implemented within 20 lines of python code. We prove that this method could be mathematically regarded as learning a Bayesian approximation of posterior attention. Experiments show that our method improved the diversity and novelty of the generated text while maintaining comparable quality on a variety of conditional and unconditional generation tasks.

Ti Wang, Hong Liu, Runwei Ding et al.

Despite substantial progress in 3D human pose estimation from a single-view image, prior works rarely explore global and local correlations, leading to insufficient learning of human skeleton representations. To address this issue, we propose a novel Interweaved Graph and Attention Network (IGANet) that allows bidirectional communications between graph convolutional networks (GCNs) and attentions. Specifically, we introduce an IGA module, where attentions are provided with local information from GCNs and GCNs are injected with global information from attentions. Additionally, we design a simple yet effective U-shaped multi-layer perceptron (uMLP), which can capture multi-granularity information for body joints. Extensive experiments on two popular benchmark datasets (i.e. Human3.6M and MPI-INF-3DHP) are conducted to evaluate our proposed method.The results show that IGANet achieves state-of-the-art performance on both datasets. Code is available at https://github.com/xiu-cs/IGANet.

Jinyi Hu, Xiaoyuan Yi, Wenhao Li et al. · tsinghua

Variational Auto-Encoder (VAE) has been widely adopted in text generation. Among many variants, recurrent VAE learns token-wise latent variables with each conditioned on the preceding ones, which captures sequential variability better in the era of RNN. However, it is unclear how to incorporate such recurrent dynamics into the recently dominant Transformer due to its parallelism. In this work, we propose TRACE, a Transformer-based recurrent VAE structure. TRACE imposes recurrence on segment-wise latent variables with arbitrarily separated text segments and constructs the posterior distribution with residual parameterization. Besides, we design an acceleration method by approximating idempotent matrices, which allows parallelism while maintaining the conditional dependence of latent variables. We demonstrate that TRACE could enhance the entanglement of each segment and preceding latent variables and deduce a non-zero lower bound of the KL term, providing a theoretical guarantee of generation diversity. Experiments on two unconditional and one conditional generation tasks show that TRACE achieves significantly improved diversity while maintaining satisfactory generation quality.

Chuyun Shen, Wenhao Li, Yuhang Shi et al.

Interactive medical image segmentation (IMIS) has shown significant potential in enhancing segmentation accuracy by integrating iterative feedback from medical professionals. However, the limited availability of enough 3D medical data restricts the generalization and robustness of most IMIS methods. The Segment Anything Model (SAM), though effective for 2D images, requires expensive semi-auto slice-by-slice annotations for 3D medical images. In this paper, we explore the zero-shot capabilities of SAM 2, the next-generation Meta SAM model trained on videos, for 3D medical image segmentation. By treating sequential 2D slices of 3D images as video frames, SAM 2 can fully automatically propagate annotations from a single frame to the entire 3D volume. We propose a practical pipeline for using SAM 2 in 3D medical image segmentation and present key findings highlighting its efficiency and potential for further optimization. Concretely, numerical experiments on the BraTS2020 and the medical segmentation decathlon datasets demonstrate that SAM 2 still has a gap with supervised methods but can narrow the gap in specific settings and organ types, significantly reducing the annotation burden on medical professionals. Our code will be open-sourced and available at https://github.com/Chuyun-Shen/SAM_2_Medical_3D.

Mind Lab, Song Cao, Vic Cao et al.

Parameter-efficient fine-tuning (PEFT) is usually treated as a cheaper alternative to full fine-tuning. We study a broader role: small trainable adapters as persistent local state on top of strong shared foundation models. In this framing, the base model provides shared competence while adapters carry instance-specific behavior such as preferences, skills, tool habits, and memory-like updates. We organize the problem around three scaling axes: Scale Up, where stronger shared priors make small local updates more useful; Scale Down, where we study how small adapters can be while remaining reliable; and Scale Out, where many persistent adapted instances coexist. MinT provides one infrastructure example for managing adapter identity, revision, provenance, evaluation, and serving residency. Together, the results suggest that PEFT can be a compact substrate for persistent personal models rather than only a budget substitute for full fine-tuning.

Wenhao Li, Xiu Su, Dan Niu et al.

Vision-Language-Action (VLA) models have demonstrated remarkable capabilities and generalization in embodied manipulation. However, their decision-making relies on a fast, instinctive process that lacks deliberation. This strategy often leads to suboptimal or catastrophic actions when facing complex or ambiguous scenarios that require greater consideration. In this paper, we introduce \textbf{VLA-ATTC}, a framework that endows VLA models with adaptive test-time compute (TTC). VLA-ATTC employs an uncertainty-based ``cognitive clutch'' to dynamically transition from reflexive execution to a TTC deliberation phase when necessary. During TTC phase, a novel \textbf{Relative Action Critic} (RAC) model identifies the optimal action from generated candidates via pairwise comparisons. This relative mechanism replaces unstable absolute value estimation, significantly simplifying the learning objective. Furthermore, we introduce an efficient sampling strategy to amortize computational costs and an automated data pipeline that curates preference pairs without manual annotation. On the LIBERO-LONG benchmark, VLA-ATTC reduces the failure rate of the SOTA model PI0.5 by over 50\%. We will open-source all the code and weights.

Tao Wang, Mengyuan Liu, Hong Liu et al.

Occluded person re-identification (Re-ID) is a challenging problem due to the destruction of occluders. Most existing methods focus on visible human body parts through some prior information. However, when complementary occlusions occur, features in occluded regions can interfere with matching, which affects performance severely. In this paper, different from most previous works that discard the occluded region, we propose a Feature Completion Transformer (FCFormer) to implicitly complement the semantic information of occluded parts in the feature space. Specifically, Occlusion Instance Augmentation (OIA) is proposed to simulates real and diverse occlusion situations on the holistic image. These augmented images not only enrich the amount of occlusion samples in the training set, but also form pairs with the holistic images. Subsequently, a dual-stream architecture with a shared encoder is proposed to learn paired discriminative features from pairs of inputs. Without additional semantic information, an occluded-holistic feature sample-label pair can be automatically created. Then, Feature Completion Decoder (FCD) is designed to complement the features of occluded regions by using learnable tokens to aggregate possible information from self-generated occluded features. Finally, we propose the Cross Hard Triplet (CHT) loss to further bridge the gap between complementing features and extracting features under the same ID. In addition, Feature Completion Consistency (FC$^2$) loss is introduced to help the generated completion feature distribution to be closer to the real holistic feature distribution. Extensive experiments over five challenging datasets demonstrate that the proposed FCFormer achieves superior performance and outperforms the state-of-the-art methods by significant margins on occluded datasets.

Jie Zhou, Xianshuai Cao, Wenhao Li et al.

Multi-scenario & multi-task learning has been widely applied to many recommendation systems in industrial applications, wherein an effective and practical approach is to carry out multi-scenario transfer learning on the basis of the Mixture-of-Expert (MoE) architecture. However, the MoE-based method, which aims to project all information in the same feature space, cannot effectively deal with the complex relationships inherent among various scenarios and tasks, resulting in unsatisfactory performance. To tackle the problem, we propose a Hierarchical information extraction Network (HiNet) for multi-scenario and multi-task recommendation, which achieves hierarchical extraction based on coarse-to-fine knowledge transfer scheme. The multiple extraction layers of the hierarchical network enable the model to enhance the capability of transferring valuable information across scenarios while preserving specific features of scenarios and tasks. Furthermore, a novel scenario-aware attentive network module is proposed to model correlations between scenarios explicitly. Comprehensive experiments conducted on real-world industrial datasets from Meituan Meishi platform demonstrate that HiNet achieves a new state-of-the-art performance and significantly outperforms existing solutions. HiNet is currently fully deployed in two scenarios and has achieved 2.87% and 1.75% order quantity gain respectively.

Wenhao Li, Xiangfeng Wang, Bo Jin et al.

Social dilemmas can be considered situations where individual rationality leads to collective irrationality. The multi-agent reinforcement learning community has leveraged ideas from social science, such as social value orientations (SVO), to solve social dilemmas in complex cooperative tasks. In this paper, by first introducing the typical "division of labor or roles" mechanism in human society, we provide a promising solution for intertemporal social dilemmas (ISD) with SVOs. A novel learning framework, called Learning Roles with Emergent SVOs (RESVO), is proposed to transform the learning of roles into the social value orientation emergence, which is symmetrically solved by endowing agents with altruism to share rewards with other agents. An SVO-based role embedding space is then constructed by individual conditioning policies on roles with a novel rank regularizer and mutual information maximizer. Experiments show that RESVO achieves a stable division of labor and cooperation in ISDs with different complexity.

Zhecan Wang, Haoxuan You, Yicheng He et al.

Visual commonsense understanding requires Vision Language (VL) models to not only understand image and text but also cross-reference in-between to fully integrate and achieve comprehension of the visual scene described. Recently, various approaches have been developed and have achieved high performance on visual commonsense benchmarks. However, it is unclear whether the models really understand the visual scene and underlying commonsense knowledge due to limited evaluation data resources. To provide an in-depth analysis, we present a Multimodal Evaluation (ME) pipeline to automatically generate question-answer pairs to test models' understanding of the visual scene, text, and related knowledge. We then take a step further to show that training with the ME data boosts the model's performance in standard VCR evaluation. Lastly, our in-depth analysis and comparison reveal interesting findings: (1) semantically low-level information can assist the learning of high-level information but not the opposite; (2) visual information is generally under utilization compared with text.

Wenhao Li, Daohai Yu, Gen Luo et al.

Training Large Language Models (LLMs) on long contexts is severely constrained by prohibitive GPU memory overhead, not training time. The primary culprits are the activations, whose memory footprints scale linearly with sequence length. We introduce OOMB, a highly memory-efficient training system that directly confronts this barrier. Our approach employs a chunk-recurrent training framework with on-the-fly activation recomputation, which maintains a constant activation memory footprint (O(1)) and shifts the primary bottleneck to the growing KV cache. To manage the KV cache, OOMB integrates a suite of synergistic optimizations: a paged memory manager for both the KV cache and its gradients to eliminate fragmentation, asynchronous CPU offloading to hide data transfer latency, and page-level sparse attention to reduce both computational complexity and communication overhead. The synergy of these techniques yields exceptional efficiency. Our empirical results show that for every additional 10K tokens of context, the end-to-end training memory overhead increases by a mere 10MB for Qwen2.5-7B. This allows training Qwen2.5-7B with a 4M-token context on a single H200 GPU, a feat that would otherwise require a large cluster using context parallelism. This work represents a substantial advance in resource efficiency for long-context LLM training. The source code is available at https://github.com/wenhaoli-xmu/OOMB.

Wenhao Li, Xiu Su, Yichao Cao et al.

Vision-language-action (VLA) models have advanced the field of embodied manipulation by harnessing broad world knowledge and strong generalization. However, current VLA models still face several key challenges, including limited reasoning capability, lack of status monitoring, and difficulty in self-correction. In this paper, we introduce \textbf{Sentinel-VLA}, a metacognitive VLA model equipped with an active ``sentinel'' module to monitor real-time execution status. Only when necessary, such as during initial planning or upon detecting an error, the model triggers a dynamic reasoning or formulate error recovery solutions. This on-demand reasoning mechanism ensures robust decision-making while minimizing computational overhead. Notably, all training data (spanning 44 tasks and over 2.6 million transitions) is automatically generated and annotated through our designed pipeline. We also propose the Self-Evolving Continual Learning (SECL) algorithm, which allows Sentinel-VLA to identify its capability boundaries and automatically collect data for expansion, paired with Orthogonal Continual Adapter (OC-Adapter) to constrain parameter updates to an orthogonal space, thereby preventing catastrophic forgetting. Real-world experiments demonstrate that Sentinel-VLA boosts the task success rate by over 30\% compared to the SOTA model, PI0. We will open-source all the code, weights, and data generation pipeline.

Junjie Sheng, Wenhao Li, Bo Jin et al.

Over-generalization is a thorny issue in cognitive science, where people may become overly cautious due to past experiences. Agents in multi-agent reinforcement learning (MARL) also have been found to suffer relative over-generalization (RO) as people do and stuck to sub-optimal cooperation. Recent methods have shown that assigning reasoning ability to agents can mitigate RO algorithmically and empirically, but there has been a lack of theoretical understanding of RO, let alone designing provably RO-free methods. This paper first proves that RO can be avoided when the MARL method satisfies a consistent reasoning requirement under certain conditions. Then we introduce a novel reasoning framework, called negotiated reasoning, that first builds the connection between reasoning and RO with theoretical justifications. After that, we propose an instantiated algorithm, Stein variational negotiated reasoning (SVNR), which uses Stein variational gradient descent to derive a negotiation policy that provably avoids RO in MARL under maximum entropy policy iteration. The method is further parameterized with neural networks for amortized learning, making computation efficient. Numerical experiments on many RO-challenged environments demonstrate the superiority and efficiency of SVNR compared to state-of-the-art methods in addressing RO.

Wenhao Li, Baoxiang Wang, Shanchao Yang et al.

Enhancing the diversity of policies is beneficial for robustness, exploration, and transfer in reinforcement learning (RL). In this paper, we aim to seek diverse policies in an under-explored setting, namely RL tasks with structured action spaces with the two properties of composability and local dependencies. The complex action structure, non-uniform reward landscape, and subtle hyperparameter tuning due to the properties of structured actions prevent existing approaches from scaling well. We propose a simple and effective RL method, Diverse Policy Optimization (DPO), to model the policies in structured action space as the energy-based models (EBM) by following the probabilistic RL framework. A recently proposed novel and powerful generative model, GFlowNet, is introduced as the efficient, diverse EBM-based policy sampler. DPO follows a joint optimization framework: the outer layer uses the diverse policies sampled by the GFlowNet to update the EBM-based policies, which supports the GFlowNet training in the inner layer. Experiments on ATSC and Battle benchmarks demonstrate that DPO can efficiently discover surprisingly diverse policies in challenging scenarios and substantially outperform existing state-of-the-art methods.

Wenhao Li, Xueying Jiang, Gongjie Zhang et al.

4D point cloud videos capture rich spatial and temporal dynamics of scenes which possess unique values in various 4D understanding tasks. However, most existing methods work in the spatiotemporal domain where the underlying geometric characteristics of 4D point cloud videos are hard to capture, leading to degraded representation learning and understanding of 4D point cloud videos. We address the above challenge from a complementary spectral perspective. By transforming 4D point cloud videos into graph spectral signals, we can decompose them into multiple frequency bands each of which captures distinct geometric structures of point cloud videos. Our spectral analysis reveals that the decomposed low-frequency signals capture more coarse shapes while high-frequency signals encode more fine-grained geometry details. Building on these observations, we design Spatio-Temporal-Spectral Mixer (STS-Mixer), a unified framework that mixes spatial, temporal, and spectral representations of point cloud videos. STS-Mixer integrates multi-band delineated spectral signals with spatiotemporal information to capture rich geometries and temporal dynamics, while enabling fine-grained and holistic understanding of 4D point cloud videos. Extensive experiments show that STS-Mixer achieves superior performance consistently across multiple widely adopted benchmarks on both 3D action recognition and 4D semantic segmentation tasks. Code and models are available at https://github.com/Vegetebird/STS-Mixer.

Yulei Ye, Wenhao Li, Zhong Wen et al.

Despite the rapid deployment of LLMs into classrooms, validating educational AI remains uniquely intractable: interventions act on developing learners whose cognitive and social trajectories are irreversibly shaped, while real-world trials are slow, ethically constrained, and institutionally locked. LLM-based educational simulators have emerged as a potential remedy, but many still collapse learning into persona-conditioned role-play and, when optimized only to reproduce existing classrooms, can structurally penalize the institutional novelty that pedagogical reform requires. In this work, we introduce AgentSchool, an LLM-driven multi-agent simulator that models learning as state transition rather than prompted behavior. AgentSchool couples cognitively growable student agents -- equipped with weighted subject knowledge graphs, thinking-workflow pools, and explicit misconceptions -- with adaptive teacher agents that plan, scaffold, and reflect along the Zone of Proximal Development, embedded in a configurable scenery generator that situates instruction within both formal and informal learning fields, and a multi-scale simulator that decouples interaction scale, temporal granularity, and simulation duration. Experiments show that structured student agents produce more differentiated mastery and misconception traces than a baseline simulator, while teacher-agent comparisons show backbone-dependent patterns consistent with ZPD-informed adaptation. Further, AgentSchool generates plausible traces of peripheral participation, clique formation, aggressor-induced cohesion, and opinion-leader emergence consistent with classroom social theories. Beyond its role as an educational research instrument, AgentSchool frames education as a socially meaningful testbed for long-horizon memory, multi-agent coordination, and future institutional reasoning under organizational pressure.

Wenwu Li, Yuran Song, Mingze Zhao et al.

While Multi-Agent Systems (MAS) empower Large Language Models to tackle complex reasoning tasks through collaborative interaction, optimizing their dynamics remains a formidable challenge due to the discrete, non-differentiable nature of the computation graph and the sparsity of global supervisory signals. Existing black-box optimizers struggle to attribute trajectory-level failure to specific local components, resulting in inefficient, high-variance exploration. We argue that tractable MAS optimization needs structural inductive biases to disentangle error signals. We propose temporal and structural credit assignment, which decomposes the objective along two axes: (i) temporal credit, using state-space bottlenecks to identify critical rounds, and (ii) structural credit, using stationary role policies to isolate agent contributions. Leveraging these decomposed signals, we introduce a discrete, verbalized block coordinate descent algorithm for iterative refinement. Rather than indiscriminate global updates, it alternates between optimizing role prompts and aggregation protocols, using LLM-generated "proxy gradients" to target only the identified weak links. Across diverse reasoning benchmarks, our approach substantially reduces query complexity while improving performance, providing a principled and interpretable path toward self-improving MAS.

Wenhao Li, Xiangfeng Wang, Bo Jin

Diffusion-based planning has achieved strong results in single-agent offline reinforcement learning, yet scaling to many-agent systems remains intractable due to the curse of dimensionality in the joint trajectory space. We introduce MF-Diffuser, a framework that lifts trajectory planning to the Wasserstein space of trajectory distributions, where the propagation of chaos ensures a small representative subset of agents captures the full population dynamics. Our approach features a value-weighted chaotic entropy objective that reconciles generative fidelity with return maximization, and a hierarchical coarse-to-fine strategy that progressively grows the agent population during denoising. We establish end-to-end suboptimality bounds with four interpretable terms, revealing that mean-field approximation error scales as $O(H^2/\sqrt{N})$ while offline distribution shift provably does not grow with population size $N$, and prove the generated policy is an approximate mean-field Nash equilibrium with explicit convergence guarantees. Experiments on three mean-field RL benchmarks -- spanning stage games, sequential dynamics, and adversarial team competition -- show MF-Diffuser achieves the best return in the majority of settings, with the largest gains on suboptimal offline data and at extreme scales ($N \geq 10^3$).

Huakang Chen, Jingbin Hu, Liumeng Xue et al.

Instruction-following text-to-speech (TTS) has emerged as an important capability for controllable and expressive speech generation, yet its evaluation remains underdeveloped due to limited benchmark coverage, weak diagnostic granularity, and insufficient multilingual support. We present \textbf{MINT-Bench}, a comprehensive multilingual benchmark for instruction-following TTS. MINT-Bench is built upon a hierarchical multi-axis taxonomy, a scalable multi-stage data construction pipeline, and a hierarchical hybrid evaluation protocol that jointly assesses content consistency, instruction following, and perceptual quality. Experiments across ten languages show that current systems remain far from solved: frontier commercial systems lead overall, while leading open-source models become highly competitive and can even outperform commercial counterparts in localized settings such as Chinese. The benchmark further reveals that harder compositional and paralinguistic controls remain major bottlenecks for current systems. We release MINT-Bench together with the data construction and evaluation toolkit to support future research on controllable, multilingual, and diagnostically grounded TTS evaluation. The leaderboard and demo are available at https://longwaytog0.github.io/MINT-Bench/

Wenhao Li, Yixuan Huang, Shengxin Zhu

This paper revisits the error analysis of the Stochastic Lanczos Quadrature (SLQ) method for approximating the trace of matrix functions, with a specific focus on asymmetric Lanczos quadrature rules. We reexplain an existing theoretical discrepancy regarding the necessity of a scaling factor when applying an affine transformation from the reference interval to the physical spectral interval. Furthermore, we introduce an optimized error reallocation technique for log-determinant estimation. Rather than evenly splitting the error tolerance between the Hutchinson trace estimator and the Lanczos quadrature, we formulate an optimization problem to strategically distribute the error budget. This approach minimizes the total number of matrix-vector multiplications (MVMs) required to reach a target accuracy for both Rademacher and Gaussian queries. Numerical experiments validate that this reallocation yields tighter theoretical bounds and provides a concrete rule-of-thumb for parameter configuration: to achieve a target accuracy efficiently, more computational resources should be allocated to the Lanczos process (larger m) rather than Monte Carlo sampling (smaller N).

Zhecan Wang, Long Chen, Haoxuan You et al.

Vision-language (VL) understanding tasks evaluate models' comprehension of complex visual scenes through multiple-choice questions. However, we have identified two dataset biases that models can exploit as shortcuts to resolve various VL tasks correctly without proper understanding. The first type of dataset bias is \emph{Unbalanced Matching} bias, where the correct answer overlaps the question and image more than the incorrect answers. The second type of dataset bias is \emph{Distractor Similarity} bias, where incorrect answers are overly dissimilar to the correct answer but significantly similar to other incorrect answers within the same sample. To address these dataset biases, we first propose Adversarial Data Synthesis (ADS) to generate synthetic training and debiased evaluation data. We then introduce Intra-sample Counterfactual Training (ICT) to assist models in utilizing the synthesized training data, particularly the counterfactual data, via focusing on intra-sample differentiation. Extensive experiments demonstrate the effectiveness of ADS and ICT in consistently improving model performance across different benchmarks, even in domain-shifted scenarios.

Wenhao Li, Zhiyuan Yu, Qijin She et al.

The household rearrangement task involves spotting misplaced objects in a scene and accommodate them with proper places. It depends both on common-sense knowledge on the objective side and human user preference on the subjective side. In achieving such task, we propose to mine object functionality with user preference alignment directly from the scene itself, without relying on human intervention. To do so, we work with scene graph representation and propose LLM-enhanced scene graph learning which transforms the input scene graph into an affordance-enhanced graph (AEG) with information-enhanced nodes and newly discovered edges (relations). In AEG, the nodes corresponding to the receptacle objects are augmented with context-induced affordance which encodes what kind of carriable objects can be placed on it. New edges are discovered with newly discovered non-local relations. With AEG, we perform task planning for scene rearrangement by detecting misplaced carriables and determining a proper placement for each of them. We test our method by implementing a tiding robot in simulator and perform evaluation on a new benchmark we build. Extensive evaluations demonstrate that our method achieves state-of-the-art performance on misplacement detection and the following rearrangement planning.

Chuyun Shen, Wenhao Li, Ya Zhang et al.

The Segmentation Anything Model (SAM) has recently emerged as a foundation model for addressing image segmentation. Owing to the intrinsic complexity of medical images and the high annotation cost, the medical image segmentation (MIS) community has been encouraged to investigate SAM's zero-shot capabilities to facilitate automatic annotation. Inspired by the extraordinary accomplishments of interactive medical image segmentation (IMIS) paradigm, this paper focuses on assessing the potential of SAM's zero-shot capabilities within the IMIS paradigm to amplify its benefits in the MIS domain. Regrettably, we observe that SAM's vulnerability to prompt forms (e.g., points, bounding boxes) becomes notably pronounced in IMIS. This leads us to develop a framework that adaptively offers suitable prompt forms for human experts. We refer to the framework above as temporally-extended prompts optimization (TEPO) and model it as a Markov decision process, solvable through reinforcement learning. Numerical experiments on the standardized benchmark BraTS2020 demonstrate that the learned TEPO agent can further enhance SAM's zero-shot capability in the MIS context.

Yingxuan You, Hong Liu, Xia Li et al.

3D human mesh recovery from a 2D pose plays an important role in various applications. However, it is hard for existing methods to simultaneously capture the multiple relations during the evolution from skeleton to mesh, including joint-joint, joint-vertex and vertex-vertex relations, which often leads to implausible results. To address this issue, we propose a novel solution, called GATOR, that contains an encoder of Graph-Aware Transformer (GAT) and a decoder with Motion-Disentangled Regression (MDR) to explore these multiple relations. Specifically, GAT combines a GCN and a graph-aware self-attention in parallel to capture physical and hidden joint-joint relations. Furthermore, MDR models joint-vertex and vertex-vertex interactions to explore joint and vertex relations. Based on the clustering characteristics of vertex offset fields, MDR regresses the vertices by composing the predicted base motions. Extensive experiments show that GATOR achieves state-of-the-art performance on two challenging benchmarks.

Hongpeng Wang, Zeyu Zhang, Wenhao Li et al.

Human motion understanding and generation are crucial for vision and robotics but remain limited in reasoning capability and test-time planning. We propose MoRL, a unified multimodal motion model trained with supervised fine-tuning and reinforcement learning with verifiable rewards. Our task-specific reward design combines semantic alignment and reasoning coherence for understanding with physical plausibility and text-motion consistency for generation, improving both logical reasoning and perceptual realism. To further enhance inference, we introduce Chain-of-Motion (CoM), a test-time reasoning method that enables step-by-step planning and reflection. We also construct two large-scale CoT datasets, MoUnd-CoT-140K and MoGen-CoT-140K, to align motion sequences with reasoning traces and action descriptions. Experiments on HumanML3D and KIT-ML show that MoRL achieves significant gains over state-of-the-art baselines. Code: https://github.com/AIGeeksGroup/MoRL. Website: https://aigeeksgroup.github.io/MoRL.

Jiehao Wu, Zixiao Huang, Wenhao Li et al.

AscendC (Ascend C) operator optimization on Huawei Ascend neural processing units (NPUs) faces a two-fold knowledge bottleneck: unlike the CUDA ecosystem, there are few public reference implementations to learn from, and performance hinges on a coupled two-part artifact - a host-side tiling program that orchestrates data movement and a kernel program that schedules and pipelines instructions. We present AscendOptimizer, an episodic agent that bootstraps this missing expertise by turning execution into experience. On the host side, AscendOptimizer performs profiling-in-the-loop evolutionary search to discover valid and high-performing tiling and data-movement configurations directly from hardware feedback. On the kernel side, it mines transferable optimization motifs by rewinding optimized kernels - systematically de-optimizing them to synthesize instructive "bad-to-good" trajectories - and distills these motifs into a retrievable experience bank for guided rewriting. By alternating host tuning and kernel rewriting in a closed loop, AscendOptimizer steadily expands feasibility and pushes latency down. On a benchmark of 127 real AscendC operators, AscendOptimizer achieves a 1.19x geometric-mean speedup over the open-source baseline, with 49.61% of operators outperforming their references, outperforming strong agent and search baselines.

Yidi Li, Jiahao Wen, Bin Ren et al.

The integration of point and voxel representations is becoming more common in LiDAR-based 3D object detection. However, this combination often struggles with capturing semantic information effectively. Moreover, relying solely on point features within regions of interest can lead to information loss and limitations in local feature representation. To tackle these challenges, we propose a novel two-stage 3D object detector, called Point-Voxel Attention Fusion Network (PVAFN). PVAFN leverages an attention mechanism to improve multi-modal feature fusion during the feature extraction phase. In the refinement stage, it utilizes a multi-pooling strategy to integrate both multi-scale and region-specific information effectively. The point-voxel attention mechanism adaptively combines point cloud and voxel-based Bird's-Eye-View (BEV) features, resulting in richer object representations that help to reduce false detections. Additionally, a multi-pooling enhancement module is introduced to boost the model's perception capabilities. This module employs cluster pooling and pyramid pooling techniques to efficiently capture key geometric details and fine-grained shape structures, thereby enhancing the integration of local and global features. Extensive experiments on the KITTI and Waymo datasets demonstrate that the proposed PVAFN achieves competitive performance. The code and models will be available.

Wenhao Li, Xiu Su, Shan You et al.

Diffusion models have recently exhibited remarkable performance on synthetic data. After a diffusion path is selected, a base model, such as UNet, operates as a denoising autoencoder, primarily predicting noises that need to be eliminated step by step. Consequently, it is crucial to employ a model that aligns with the expected budgets to facilitate superior synthetic performance. In this paper, we meticulously analyze the diffusion model and engineer a base model search approach, denoted "DiffNAS". Specifically, we leverage GPT-4 as a supernet to expedite the search, supplemented with a search memory to enhance the results. Moreover, we employ RFID as a proxy to promptly rank the experimental outcomes produced by GPT-4. We also adopt a rapid-convergence training strategy to boost search efficiency. Rigorous experimentation corroborates that our algorithm can augment the search efficiency by 2 times under GPT-based scenarios, while also attaining a performance of 2.82 with 0.37 improvement in FID on CIFAR10 relative to the benchmark IDDPM algorithm.

Zhecan Wang, Junzhang Liu, Chia-Wei Tang et al.

Existing vision-language understanding benchmarks largely consist of images of objects in their usual contexts. As a consequence, recent multimodal large language models can perform well with only a shallow visual understanding by relying on background language biases. Thus, strong performance on these benchmarks does not necessarily correlate with strong visual understanding. In this paper, we release JourneyBench, a comprehensive human-annotated benchmark of generated images designed to assess the model's fine-grained multimodal reasoning abilities across five tasks: complementary multimodal chain of thought, multi-image VQA, imaginary image captioning, VQA with hallucination triggers, and fine-grained retrieval with sample-specific distractors. Unlike existing benchmarks, JourneyBench explicitly requires fine-grained multimodal reasoning in unusual imaginary scenarios where language bias and holistic image gist are insufficient. We benchmark state-of-the-art models on JourneyBench and analyze performance along a number of fine-grained dimensions. Results across all five tasks show that JourneyBench is exceptionally challenging for even the best models, indicating that models' visual reasoning abilities are not as strong as they first appear. We discuss the implications of our findings and propose avenues for further research.

Wenhao Li, Yu Nie, Zhongyao Yang et al.

In this paper, the influence of fan-shaped buffer zone on the performance of the toll plaza is researched. A two-dimensional traffic flow model and a comprehensive evaluation model based on mechanical model and psychological field are established. The traffic flow model is simulated by creating coordinate system. We first establish queue theory model to analyze vehicles when entering toll plaza. Then, a two-dimensional steadily car-following model is established based on psychological field for the analysis of vehicles when leaving toll plaza. According to psychological field theory, we analyze the force condition of each vehicle. The force of each vehicle is contributed by the vehicles in its observation area and obstacles. By projecting these vehicles and obstacles via the equipotential line in the psychological field, the influence on the value and direction acceleration of following vehicles is obtained. Consequently, the changes of each vehicle's speed and position are obtained as well. Next, we establish simulation based on the states of vehicles and make the rules of vehicle state-changing. By simulating the system, we obtain the throughput of the toll plaza's input and output. Then we obtained the bearing pressure on the road by the max throughput and the demand of the roads. Using the number of cars in per unit area as the safety factor. Then a comprehensive evaluation model is established based on bearing pressure on the road, cost and safety factor.

Yudong W. Xu, Wenhao Li, Xiaoyu Wang et al.

Diffusion models have shown promise in learning to solve constraint optimization problems. However, they are mostly restricted to problems with binary variables and rely on graph neural networks, hindering their application to a broader range of problems such as those with general discrete variables or constraint structures that necessitate global rather than local reasoning. We investigate the use of Diffusion Transformers to address the aforementioned limitations. A naive implementation performs poorly due to a fundamental mismatch between the standard diffusion process and constraint solving: while the former applies small, incremental denoising across all variables, the latter requires substantially altering specific subsets of variables to attain feasibility or optimality. Our method, Blocked Gibbs Diffusion Transformer (BloGDiT), is the first to address this limitation by replacing standard joint Gaussian denoising with blocked Gaussian denoising. BloGDiT uses iterative block resampling and anneals the block size over time to facilitate large, targeted edits within a block of variables. Across Sudoku, Graph Coloring, Maximum Independent Set, and MaxCut, BloGDiT matches or outperforms existing methods, demonstrating that blocked Gibbs-style diffusion provides a highly effective inductive bias for Transformer-based constraint satisfaction and optimization.

Weiyun Wang, Zhangwei Gao, Lixin Gu et al. · cmu, pku

We introduce InternVL 3.5, a new family of open-source multimodal models that significantly advances versatility, reasoning capability, and inference efficiency along the InternVL series. A key innovation is the Cascade Reinforcement Learning (Cascade RL) framework, which enhances reasoning through a two-stage process: offline RL for stable convergence and online RL for refined alignment. This coarse-to-fine training strategy leads to substantial improvements on downstream reasoning tasks, e.g., MMMU and MathVista. To optimize efficiency, we propose a Visual Resolution Router (ViR) that dynamically adjusts the resolution of visual tokens without compromising performance. Coupled with ViR, our Decoupled Vision-Language Deployment (DvD) strategy separates the vision encoder and language model across different GPUs, effectively balancing computational load. These contributions collectively enable InternVL3.5 to achieve up to a +16.0\% gain in overall reasoning performance and a 4.05$\times$ inference speedup compared to its predecessor, i.e., InternVL3. In addition, InternVL3.5 supports novel capabilities such as GUI interaction and embodied agency. Notably, our largest model, i.e., InternVL3.5-241B-A28B, attains state-of-the-art results among open-source MLLMs across general multimodal, reasoning, text, and agentic tasks -- narrowing the performance gap with leading commercial models like GPT-5. All models and code are publicly released.

Yudong W. Xu, Wenhao Li, Scott Sanner et al. · utoronto

Neural networks are being increasingly used as heuristics for constraint satisfaction. These neural methods are often recurrent, learning to iteratively refine candidate assignments. In this work, we make explicit the connection between such iterative neural heuristics and Large Neighborhood Search (LNS), and adapt an existing neural constraint satisfaction method-ConsFormer-into an LNS procedure. We decompose the resulting neural LNS into two standard components: the destroy and repair operators. On the destroy side, we instantiate several classical heuristics and introduce novel prediction-guided operators that exploit the model's internal scores to select neighborhoods. On the repair side, we utilize ConsFormer as a neural repair operator and compare the original sampling-based decoder to a greedy decoder that selects the most likely assignments. Through an empirical study on Sudoku, Graph Coloring, and MaxCut, we find that adapting the neural heuristic to an LNS procedure yields substantial gains over its vanilla settings and improves its competitiveness with classical and neural baselines. We further observe consistent design patterns across tasks: stochastic destroy operators outperform greedy ones, while greedy repair is more effective than sampling-based repair for finding a single high-quality feasible assignment. These findings highlight LNS as a useful lens and design framework for structuring and improving iterative neural approaches.

Ningyuan Chen, Ming Hu, Wenhao Li

Commercial AI solutions provide analysts and managers with data-driven business intelligence for a wide range of decisions, such as demand forecasting and pricing. However, human analysts may have their own insights and experiences about the decision-making that is at odds with the algorithmic recommendation. In view of such a conflict, we provide a general analytical framework to study the augmentation of algorithmic decisions with human knowledge: the analyst uses the knowledge to set a guardrail by which the algorithmic decision is clipped if the algorithmic output is out of bound, and seems unreasonable. We study the conditions under which the augmentation is beneficial relative to the raw algorithmic decision. We show that when the algorithmic decision is asymptotically optimal with large data, the non-data-driven human guardrail usually provides no benefit. However, we point out three common pitfalls of the algorithmic decision: (1) lack of domain knowledge, such as the market competition, (2) model misspecification, and (3) data contamination. In these cases, even with sufficient data, the augmentation from human knowledge can still improve the performance of the algorithmic decision.

Wenhao Li, Yichao Cao, Xiu Su et al.

Video generation models hold substantial potential in areas such as filmmaking. However, current video diffusion models need high computational costs and produce suboptimal results due to extreme complexity of video generation task. In this paper, we propose \textbf{ConFiner}, an efficient video generation framework that decouples video generation into easier subtasks: structure \textbf{con}trol and spatial-temporal re\textbf{fine}ment. It can generate high-quality videos with chain of off-the-shelf diffusion model experts, each expert responsible for a decoupled subtask. During the refinement, we introduce coordinated denoising, which can merge multiple diffusion experts' capabilities into a single sampling. Furthermore, we design ConFiner-Long framework, which can generate long coherent video with three constraint strategies on ConFiner. Experimental results indicate that with only 10\% of the inference cost, our ConFiner surpasses representative models like Lavie and Modelscope across all objective and subjective metrics. And ConFiner-Long can generate high-quality and coherent videos with up to 600 frames.

Xueying Jiang, Wenhao Li, Quanhao Qian et al.

3D localization in Multimodal Large Language Models (MLLMs), including 3D object detection and 3D visual grounding, is fundamentally limited by camera intrinsic ambiguity: the same image admits different 3D scenes under different cameras. Existing MLLMs either ignore camera parameters and overfit to a canonical training intrinsic, or retrieve depth and 3D cues from external tools but treat the returned values as reference cues (numerical hints that the model is free to interpret implicitly), both preventing camera information from being deterministically propagated into the prediction. We propose an equation-anchored tool-use framework that re-purposes spatial tools as formula variables. The proposed framework proactively retrieves camera intrinsics and samples multi-point metric depths, writes the pinhole back-projection equation $\hat{X} = (u_c - c_x)\bar{Z}/f_x$ explicitly in Chain-of-Thought (CoT), and substitutes tool outputs into the formula before regressing the final 9-DoF bounding box. On both 3D object detection and 3D visual grounding tasks under rescaled camera intrinsics from $0.5\times$ to $1.5\times$, our method outperforms RGB-only and tool-augmented baselines, with significant gains where the camera deviates most from the training scale. Code and data will be released.

MiniCPM Team, Chaojun Xiao, Yuxuan Li et al. · tencent-ai, tsinghua

This paper introduces MiniCPM4, a highly efficient large language model (LLM) designed explicitly for end-side devices. We achieve this efficiency through systematic innovation in four key dimensions: model architecture, training data, training algorithms, and inference systems. Specifically, in terms of model architecture, we propose InfLLM v2, a trainable sparse attention mechanism that accelerates both prefilling and decoding phases for long-context processing. Regarding training data, we propose UltraClean, an efficient and accurate pre-training data filtering and generation strategy, and UltraChat v2, a comprehensive supervised fine-tuning dataset. These datasets enable satisfactory model performance to be achieved using just 8 trillion training tokens. Regarding training algorithms, we propose ModelTunnel v2 for efficient pre-training strategy search, and improve existing post-training methods by introducing chunk-wise rollout for load-balanced reinforcement learning and data-efficient tenary LLM, BitCPM. Regarding inference systems, we propose CPM.cu that integrates sparse attention, model quantization, and speculative sampling to achieve efficient prefilling and decoding. To meet diverse on-device requirements, MiniCPM4 is available in two versions, with 0.5B and 8B parameters, respectively. Furthermore, we construct a hybrid reasoning model, MiniCPM4.1, which can be used in both deep reasoning mode and non-reasoning mode. Evaluation results demonstrate that MiniCPM4 and MiniCPM4.1 outperform similar-sized open-source models across benchmarks, with the 8B variants showing significant speed improvements on long sequence understanding and generation.

Yun Hua, Shang Gao, Wenhao Li et al.

In multi-agent reinforcement learning, each agent acts to maximize its individual accumulated rewards. Nevertheless, individual accumulated rewards could not fully reflect how others perceive them, resulting in selfish behaviors that undermine global performance. The externality theory, defined as ``the activities of one economic actor affect the activities of another in ways that are not reflected in market transactions,'' is applicable to analyze the social dilemmas in MARL. One of its most profound non-market solutions, ``Pigovian Tax'', which internalizes externalities by taxing those who create negative externalities and subsidizing those who create positive externalities, could aid in developing a mechanism to resolve MARL's social dilemmas. The purpose of this paper is to apply externality theory to analyze social dilemmas in MARL. To internalize the externalities in MARL, the \textbf{L}earning \textbf{O}ptimal \textbf{P}igovian \textbf{T}ax method (LOPT), is proposed, where an additional agent is introduced to learn the tax/allowance allocation policy so as to approximate the optimal ``Pigovian Tax'' which accurately reflects the externalities for all agents. Furthermore, a reward shaping mechanism based on the approximated optimal ``Pigovian Tax'' is applied to reduce the social cost of each agent and tries to alleviate the social dilemmas. Compared with existing state-of-the-art methods, the proposed LOPT leads to higher collective social welfare in both the Escape Room and the Cleanup environments, which shows the superiority of our method in solving social dilemmas.

Ningyuan Chen, Wenhao Li

We consider a decision maker allocating one unit of renewable and divisible resource in each period on a number of arms. The arms have unknown and random rewards whose means are proportional to the allocated resource and whose variances are proportional to an order $b$ of the allocated resource. In particular, if the decision maker allocates resource $A_i$ to arm $i$ in a period, then the reward $Y_i$ is$Y_i(A_i)=A_i μ_i+A_i^b ξ_{i}$, where $μ_i$ is the unknown mean and the noise $ξ_{i}$ is independent and sub-Gaussian. When the order $b$ ranges from 0 to 1, the framework smoothly bridges the standard stochastic multi-armed bandit and online learning with full feedback. We design two algorithms that attain the optimal gap-dependent and gap-independent regret bounds for $b\in [0,1]$, and demonstrate a phase transition at $b=1/2$. The theoretical results hinge on a novel concentration inequality we have developed that bounds a linear combination of sub-Gaussian random variables whose weights are fractional, adapted to the filtration, and monotonic.

Di Wu, Jiaxin Fan, Junzhe Zang et al.

Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and natural language goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle with the temporal reasoning, spatial understanding, and commonsense grounding needed for planning in interactive environments. In this work, we introduce a reinforcement fine-tuning framework that brings R1-style reasoning enhancement into embodied planning. We first distill a high-quality dataset from a powerful closed-source model and perform supervised fine-tuning (SFT) to equip the model with structured decision-making priors. We then design a rule-based reward function tailored to multi-step action quality and optimize the policy via Generalized Reinforced Preference Optimization (GRPO). Our approach is evaluated on Embench, a recent benchmark for interactive embodied tasks, covering both in-domain and out-of-domain scenarios. Experimental results show that our method significantly outperforms models of similar or larger scale, including GPT-4o-mini and 70B+ open-source baselines, and exhibits strong generalization to unseen environments. This work highlights the potential of reinforcement-driven reasoning to advance long-horizon planning in embodied AI.

Mingda Han, Huanqi Yang, Zehua Sun et al.

Millimeter-wave (mmWave) radar enables privacy-preserving human activity recognition (HAR), yet real-world deployment remains hindered by costly annotation and poor transferability under domain shift. Although prior efforts partially alleviate these challenges, most still require retraining or adaptation for each new deployment setting. This keeps mmWave HAR in a repeated collect-tune-redeploy cycle, making scalable real-world deployment difficult. In this paper, we present RAGent, a deployment-time training-free framework for mmWave HAR that reformulates recognition as evidence-grounded inference over reusable radar knowledge rather than deployment-specific model optimization. Offline, RAGent constructs a reusable radar knowledge base through constrained cross-modal supervision, where a Vision-Language Model (VLM) transfers activity semantics from synchronized videos to paired radar segments without manual radar annotation. At deployment time, RAGent recognizes activities from radar alone by retrieving physically comparable precedents in an explicit kinematic space and resolving the final label through structured multi-role reasoning. The reasoning protocol is further refined offline through zero-gradient self-evolution. Extensive experiments on a self-collected dataset show that RAGent achieves 93.39% accuracy without per-domain retraining or target-domain adaptation, while generalizing robustly across domains.