Hefeng Zhou

Hongze Liu, Chang Guo, Yingzeng Li et al.

The transition to open, distributed Multi-Agent Systems (MAS) promises scalable intelligence but introduces a non-trivial tension: maximizing global efficiency requires cooperative, resource-aware scheduling, yet autonomous agents may be self-interested and cannot be managed by a centralized controller. Prior approaches fall short in two key areas: they typically focus on single-query routing, neglecting long-term resource reuse (e.g., KV-caching) and the complexities of system-level many-to-many matching; furthermore, they rely on generic incentive mechanisms that ignore the distinct characteristics of LLM inference. To bridge this gap, we propose IEMAS (Incentive-Efficiency Mechanism for Multi-Agent Systems), a distributed framework that aligns economic incentives with system performance. IEMAS integrates a probabilistic predictive model to estimate Quality of Service (QoS) under uncertainty, which feeds into a VCG-based bipartite matching mechanism. This design guarantees truthful capability reporting and social optimality while explicitly leveraging KV cache-affinity to minimize computational redundancy. We implement IEMAS on top of vLLM and evaluate it via extensive simulations. Results demonstrate that our incentive-efficiency co-design reducing average service cost by 35% and end-to-end latency by up to 2.9 compared to baselines.

Hefeng Zhou, Xuan Liu, Sicheng Chen et al.

Federated cross-modal retrieval faces severe challenges from heterogeneous client data, particularly non-IID semantic distributions and missing modalities. Under such heterogeneity, a single global model is often insufficient to capture both shared cross-modal knowledge and client-specific characteristics. We propose RCSR, a personalization-friendly federated framework that integrates prototype anchoring, retrieval-centric semantic routing, and optional client-specific adapters. Built on a frozen CLIP backbone, RCSR leverages lightweight shared adapters for global knowledge transfer while supporting efficient local personalization. Prototype anchoring helps unimodal clients align with global cross-modal semantics, and a server-side semantic router adaptively assigns aggregation weights based on retrieval consistency to mitigate alignment drift during heterogeneous updates. Extensive experiments on MS-COCO, Flickr30K, and other benchmarks show that RCSR consistently improves global retrieval accuracy and training stability, while further enhancing client-level retrieval performance, especially for clients with incomplete modalities. Code is available at https://github.com/RezinChow/RCSR-Retrieval-Centric-Semantic-Routing.

Kai Wang, Biaojie Zeng, Zeming Wei et al.

With the rapid development of LLM-based multi-agent systems (MAS), their significant safety and security concerns have emerged, which introduce novel risks going beyond single agents or LLMs. Despite attempts to address these issues, the existing literature lacks a cohesive safeguarding system specialized for MAS risks. In this work, we introduce TrinityGuard, a comprehensive safety evaluation and monitoring framework for LLM-based MAS, grounded in the OWASP standards. Specifically, TrinityGuard encompasses a three-tier fine-grained risk taxonomy that identifies 20 risk types, covering single-agent vulnerabilities, inter-agent communication threats, and system-level emergent hazards. Designed for scalability across various MAS structures and platforms, TrinityGuard is organized in a trinity manner, involving an MAS abstraction layer that can be adapted to any MAS structures, an evaluation layer containing risk-specific test modules, alongside runtime monitor agents coordinated by a unified LLM Judge Factory. During Evaluation, TrinityGuard executes curated attack probes to generate detailed vulnerability reports for each risk type, where monitor agents analyze structured execution traces and issue real-time alerts, enabling both pre-development evaluation and runtime monitoring. We further formalize these safety metrics and present detailed case studies across various representative MAS examples, showcasing the versatility and reliability of TrinityGuard. Overall, TrinityGuard acts as a comprehensive framework for evaluating and monitoring various risks in MAS, paving the way for further research into their safety and security.

Qiang Li, Yufeng Wu, Zhan Xu et al.

In contemporary society, the escalating pressures of life and work have propelled psychological disorders to the forefront of modern health concerns, an issue that has been further accentuated by the COVID-19 pandemic. The prevalence of depression among adolescents is steadily increasing, and traditional diagnostic methods, which rely on scales or interviews, prove particularly inadequate for detecting depression in young people. Addressing these challenges, numerous AI-based methods for assisting in the diagnosis of mental health issues have emerged. However, most of these methods center around fundamental issues with scales or use multimodal approaches like facial expression recognition. Diagnosis of depression risk based on everyday habits and behaviors has been limited to small-scale qualitative studies. Our research leverages adolescent census data to predict depression risk, focusing on children's experiences with depression and their daily life situations. We introduced a method for managing severely imbalanced high-dimensional data and an adaptive predictive approach tailored to data structure characteristics. Furthermore, we proposed a cloud-based architecture for automatic online learning and data updates. This study utilized publicly available NSCH youth census data from 2020 to 2022, encompassing nearly 150,000 data entries. We conducted basic data analyses and predictive experiments, demonstrating significant performance improvements over standard machine learning and deep learning algorithms. This affirmed our data processing method's broad applicability in handling imbalanced medical data. Diverging from typical predictive method research, our study presents a comprehensive architectural solution, considering a wider array of user needs.

Honghong Zeng, Jiong Lou, Zhe Wang et al.

Prototype-based federated learning (PFL) has emerged as a promising paradigm to address data heterogeneity problems in federated learning, as it leverages mean feature vectors as prototypes to enhance model generalization. However, its robustness against backdoor attacks remains largely unexplored. In this paper, we identify that PFL is inherently resistant to existing backdoor attacks due to its unique prototype learning mechanism and local data heterogeneity. To further explore the security of PFL, we propose BAPFL, the first backdoor attack method specifically designed for PFL frameworks. BAPFL integrates a prototype poisoning strategy with a trigger optimization mechanism. The prototype poisoning strategy manipulates the trajectories of global prototypes to mislead the prototype training of benign clients, pushing their local prototypes of clean samples away from the prototypes of trigger-embedded samples. Meanwhile, the trigger optimization mechanism learns a unique and stealthy trigger for each potential target label, and guides the prototypes of trigger-embedded samples to align closely with the global prototype of the target label. Experimental results across multiple datasets and PFL variants demonstrate that BAPFL achieves a 35\%-75\% improvement in attack success rate compared to traditional backdoor attacks, while preserving main task accuracy. These results highlight the effectiveness, stealthiness, and adaptability of BAPFL in PFL.

Shanghai AI Lab, Yicheng Bao, Guanxu Chen et al.

We introduce SafeWork-R1, a cutting-edge multimodal reasoning model that demonstrates the coevolution of capabilities and safety. It is developed by our proposed SafeLadder framework, which incorporates large-scale, progressive, safety-oriented reinforcement learning post-training, supported by a suite of multi-principled verifiers. Unlike previous alignment methods such as RLHF that simply learn human preferences, SafeLadder enables SafeWork-R1 to develop intrinsic safety reasoning and self-reflection abilities, giving rise to safety `aha' moments. Notably, SafeWork-R1 achieves an average improvement of $46.54\%$ over its base model Qwen2.5-VL-72B on safety-related benchmarks without compromising general capabilities, and delivers state-of-the-art safety performance compared to leading proprietary models such as GPT-4.1 and Claude Opus 4. To further bolster its reliability, we implement two distinct inference-time intervention methods and a deliberative search mechanism, enforcing step-level verification. Finally, we further develop SafeWork-R1-InternVL3-78B, SafeWork-R1-DeepSeek-70B, and SafeWork-R1-Qwen2.5VL-7B. All resulting models demonstrate that safety and capability can co-evolve synergistically, highlighting the generalizability of our framework in building robust, reliable, and trustworthy general-purpose AI.

Yuxin Liu, Yuezhang Peng, Hefeng Zhou et al.

The widespread deployment of cameras has led to an exponential increase in video data, creating vast opportunities for applications such as traffic management and crime surveillance. However, querying specific objects from large-scale video datasets presents challenges, including (1) processing massive and continuously growing data volumes, (2) supporting complex query requirements, and (3) ensuring low-latency execution. Existing video analysis methods struggle with either limited adaptability to unseen object classes or suffer from high query latency. In this paper, we present LOVO, a novel system designed to efficiently handle comp$\underline{L}$ex $\underline{O}$bject queries in large-scale $\underline{V}$ide$\underline{O}$ datasets. Agnostic to user queries, LOVO performs one-time feature extraction using pre-trained visual encoders, generating compact visual embeddings for key frames to build an efficient index. These visual embeddings, along with associated bounding boxes, are organized in an inverted multi-index structure within a vector database, which supports queries for any objects. During the query phase, LOVO transforms object queries to query embeddings and conducts fast approximate nearest-neighbor searches on the visual embeddings. Finally, a cross-modal rerank is performed to refine the results by fusing visual features with detailed textual features. Evaluation on real-world video datasets demonstrates that LOVO outperforms existing methods in handling complex queries, with near-optimal query accuracy and up to 85x lower search latency, while significantly reducing index construction costs. This system redefines the state-of-the-art object query approaches in video analysis, setting a new benchmark for complex object queries with a novel, scalable, and efficient approach that excels in dynamic environments.

Jun Wang, Hefeng Zhou, Xiaohan Yu

Current state-of-the-art weakly supervised object detection (WSOD) studies mainly follow a two-stage training strategy which integrates a fully supervised detector (FSD) with a pure WSOD model. There are two main problems hindering the performance of the two-phase WSOD approaches, i.e., insufficient learning problem and strict reliance between the FSD and the pseudo ground truth (PGT) generated by the WSOD model. This paper proposes pseudo ground truth refinement network (PGTRNet), a simple yet effective method without introducing any extra learnable parameters, to cope with these problems. PGTRNet utilizes multiple bounding boxes to establish the PGT, mitigating the insufficient learning problem. Besides, we propose a novel online PGT refinement approach to steadily improve the quality of PGT by fully taking advantage of the power of FSD during the second-phase training, decoupling the first and second-phase models. Elaborate experiments are conducted on the PASCAL VOC 2007 benchmark to verify the effectiveness of our methods. Experimental results demonstrate that PGTRNet boosts the backbone model by 2.1% mAP and achieves the state-of-the-art performance.

Jun Wang, Qianying Liu, Haotian Xie et al.

In recent years, advances in the development of whole-slide images have laid a foundation for the utilization of digital images in pathology. With the assistance of computer images analysis that automatically identifies tissue or cell types, they have greatly improved the histopathologic interpretation and diagnosis accuracy. In this paper, the Convolutional Neutral Network (CNN) has been adapted to predict and classify lymph node metastasis in breast cancer. Unlike traditional image cropping methods that are only suitable for large resolution images, we propose a novel data augmentation method named Random Center Cropping (RCC) to facilitate small resolution images. RCC enriches the datasets while retaining the image resolution and the center area of images. In addition, we reduce the downsampling scale of the network to further facilitate small resolution images better. Moreover, Attention and Feature Fusion (FF) mechanisms are employed to improve the semantic information of images. Experiments demonstrate that our methods boost performances of basic CNN architectures. And the best-performed method achieves an accuracy of 97.96% and an AUC of 99.68% on RPCam datasets, respectively.