Wei Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Wei Wu has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Hui Xiong
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China and The Hong Kong University of Science and Technology, Hong Kong SAR, China
3 shared publications - Liyi Chen
Xiaohongshu Inc., Afghanistan
2 shared publications Mingyang Li
Alibaba Cloud Computing
2 shared publicationsMingze Yin
College of Computer Science and Technology Zhejiang University
2 shared publicationsZheng Wang
Alibaba Cloud Computing
2 shared publications- Chao Wang
School of Artificial Intelligence and Data Science, University of Science and Technology of China, Hefei, China
1 shared publication
Latest Publications
- Top Cited PapersOpen Access
- 28 April 2026
- Published by MDPI AG in Preprints.org
Agent Harness for Large Language Model Agents: A Survey
Nowadays, the reliability of large language model (LLM) agents in production environments is increasingly determined not by the underlying model but by the agent harness that encapsulates it. As tasks grow longer and more complex, recent studies demonstrate order-of-magnitude reliability gains achieved through harness redesign alone, with the underlying model held fixed. However, existing literature typically focuses on individual components of harness such as memory, planning, and tool use in isolation. In this paper, we are the first to conduct a systematic survey on the LLM agent harness, based on a comprehensive review of over hundred papers, technical reports, and industry blogs. Specifically, the core works of this paper is presented below: (1) A formal definition of the agent harness as a six-component tuple H = (E,T,C,S,L,V), i.e., execution loop, tool registry, context manager, state store, lifecycle hooks, and evaluation interface. (2) A historical tracing the evolution of the harness concept from software testing and reinforcement learning environments to modern LLM agent systems, identifying a unifying architectural pattern: enabling a controllable, observable, verifiable runtime environment for unpredictable execution agents. (3) An empirically grounded taxonomy of 23 representative systems, classified via a six-component completeness matrix that enables direct cross-framework comparison, and reveal systems mature enough for real-world production deployment consistently exhibit a complete implementation of all six architectural components. (4) A systematic analysis of nine cross-cutting technical challenges spanning sandboxing, evaluation, protocol standardization, and compute economics, including an empirical comparison of tool-level and agent-level interoperability protocols and an assessment of ultra-long-context model implications. (5) A proposal of key future research directions where harness-layer infrastructure remains significantly underdeveloped relative to advances in component capabilities. The latest version of this paper is available at: https://github.com/Gloriaameng/Awesome-Agent-Harness
- 28 April 2026
- Published by MDPI AG in Preprints.org
Top Cited PapersOpen AccessAgent Harness for Large Language Model Agents: A Survey
Nowadays, the reliability of large language model (LLM) agents in production environments is increasingly determined not by the underlying model but by the agent harness that encapsulates it. As tasks grow longer and more complex, recent studies demonstrate order-of-magnitude reliability gains achieved through harness redesign alone, with the underlying model held fixed. However, existing literature typically focuses on individual components of harness such as memory, planning, and tool use in isolation. In this paper, we are the first to conduct a systematic survey on the LLM agent harness, based on a comprehensive review of over hundred papers, technical reports, and industry blogs. Specifically, the core works of this paper is presented below: (1) A formal definition of the agent harness as a six-component tuple H = (E,T,C,S,L,V), i.e., execution loop, tool registry, context manager, state store, lifecycle hooks, and evaluation interface. (2) A historical tracing the evolution of the harness concept from software testing and reinforcement learning environments to modern LLM agent systems, identifying a unifying architectural pattern: enabling a controllable, observable, verifiable runtime environment for unpredictable execution agents. (3) An empirically grounded taxonomy of 23 representative systems, classified via a six-component completeness matrix that enables direct cross-framework comparison, and reveal systems mature enough for real-world production deployment consistently exhibit a complete implementation of all six architectural components. (4) A systematic analysis of nine cross-cutting technical challenges spanning sandboxing, evaluation, protocol standardization, and compute economics, including an empirical comparison of tool-level and agent-level interoperability protocols and an assessment of ultra-long-context model implications. (5) A proposal of key future research directions where harness-layer infrastructure remains significantly underdeveloped relative to advances in component capabilities. The latest version of this paper is available at: https://github.com/Gloriaameng/Awesome-Agent-Harness
- Open Access
Turning Semantics into Topology: LLM-Driven Attribute Augmentation for Collaborative Filtering
Large Language Models (LLMs) have shown great potential for enhancing recommender systems through their extensive world knowledge and reasoning capabilities. However, effectively translating these semantic signals into traditional collaborative embeddings remains an open challenge. Existing approaches typically fall into two extremes: direct inference methods are computationally prohibitive for large-scale retrieval, while embedding-based methods primarily focus on unilateral feature augmentation rather than holistic collaborative signal enhancement. To bridge this gap, we propose Topology-Augmented Graph Collaborative Filtering (TAGCF), a novel framework that transforms semantic knowledge into topological connectivity. Unlike existing approaches that depend on textual features or direct interaction synthesis, TAGCF employs LLMs to infer interaction intents and underlying causal relationships from user-item pairs, representing these insights as intermediate attribute nodes within an enriched User-Attribute-Item (U-A-I) graph. Furthermore, to effectively model the heterogeneous relations in this augmented structure, we propose Adaptive Relation-weighted Graph Convolution (ARGC), which employs relation-specific prediction networks to dynamically estimate the importance of each relation type. Extensive experiments across multiple benchmark datasets and CF backbones demonstrate consistent improvements, with comprehensive evaluations including cold-start scenarios validating the effectiveness and robustness of our framework. All code will be made publicly available. For anonymous review, our code is available at the following anonymous link: https://anonymous.4open.science/r/AGCF-2441353190/.
Open AccessTurning Semantics into Topology: LLM-Driven Attribute Augmentation for Collaborative Filtering
Large Language Models (LLMs) have shown great potential for enhancing recommender systems through their extensive world knowledge and reasoning capabilities. However, effectively translating these semantic signals into traditional collaborative embeddings remains an open challenge. Existing approaches typically fall into two extremes: direct inference methods are computationally prohibitive for large-scale retrieval, while embedding-based methods primarily focus on unilateral feature augmentation rather than holistic collaborative signal enhancement. To bridge this gap, we propose Topology-Augmented Graph Collaborative Filtering (TAGCF), a novel framework that transforms semantic knowledge into topological connectivity. Unlike existing approaches that depend on textual features or direct interaction synthesis, TAGCF employs LLMs to infer interaction intents and underlying causal relationships from user-item pairs, representing these insights as intermediate attribute nodes within an enriched User-Attribute-Item (U-A-I) graph. Furthermore, to effectively model the heterogeneous relations in this augmented structure, we propose Adaptive Relation-weighted Graph Convolution (ARGC), which employs relation-specific prediction networks to dynamically estimate the importance of each relation type. Extensive experiments across multiple benchmark datasets and CF backbones demonstrate consistent improvements, with comprehensive evaluations including cold-start scenarios validating the effectiveness and robustness of our framework. All code will be made publicly available. For anonymous review, our code is available at the following anonymous link: https://anonymous.4open.science/r/AGCF-2441353190/.
- Open Access
Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.
Open AccessAnti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models
Large reasoning models enhanced by reinforcement learning with verifiable rewards have achieved significant performance gains by extending their chain-of-thought. However, this paradigm incurs substantial deployment costs as models often exhibit excessive verbosity on simple queries. Existing efficient reasoning methods relying on explicit length penalties often introduce optimization conflicts and leave the generative mechanisms driving overthinking largely unexamined. In this paper, we identify a phenomenon termed length shift where models increasingly generate unnecessary reasoning on trivial inputs during training. To address this, we introduce Dynamic Outlier Truncation (DOT), a training-time intervention that selectively suppresses redundant tokens. This method targets only the extreme tail of response lengths within fully correct rollout groups while preserving long-horizon reasoning capabilities for complex problems. To complement this intervention and ensure stable convergence, we further incorporate auxiliary KL regularization and predictive dynamic sampling. Experimental results across multiple model scales demonstrate that our approach significantly pushes the efficiency-performance Pareto frontier outward. Notably, on the AIME-24, our method reduces inference token usage by 78% while simultaneously increasing accuracy compared to the initial policy and surpassing state-of-the-art efficient reasoning methods.
- Open Access
ProtCLIP: Function-Informed Protein Multi-Modal Learning
Multi-modality pre-training paradigm that aligns protein sequences and biological descriptions has learned general protein representations and achieved promising performance in various downstream applications. However, these works were still unable to replicate the extraordinary success of language-supervised visual foundation models due to the ineffective usage of aligned protein-text paired data and the lack of an effective function-informed pre-training paradigm. To address these issues, this paper curates a large-scale protein-text paired dataset called ProtAnno with a property-driven sampling strategy, and introduces a novel function-informed protein pre-training paradigm. Specifically, the sampling strategy determines selecting probability based on the sample confidence and property coverage, balancing the data quality and data quantity in face of large-scale noisy data. Furthermore, motivated by significance of the protein specific functional mechanism, the proposed paradigm explicitly model protein static and dynamic functional segments by two segment-wise pre-training objectives, injecting fine-grained information in a function-informed manner. Leveraging all these innovations, we develop ProtCLIP, a multi-modality foundation model that comprehensively represents function-aware protein embeddings. On 22 different protein benchmarks within 5 types, including protein functionality classification, mutation effect prediction, cross-modal transformation, semantic similarity inference and protein-protein interaction prediction, our ProtCLIP consistently achieves SOTA performance, with remarkable improvements of 75% on average in five cross-modal transformation benchmarks, 59.9% in GO-CC and 39.7% in GO-BP protein function prediction. The experimental results verify the extraordinary potential of ProtCLIP serving as the protein multi-modality foundation model.
Open AccessProtCLIP: Function-Informed Protein Multi-Modal Learning
Multi-modality pre-training paradigm that aligns protein sequences and biological descriptions has learned general protein representations and achieved promising performance in various downstream applications. However, these works were still unable to replicate the extraordinary success of language-supervised visual foundation models due to the ineffective usage of aligned protein-text paired data and the lack of an effective function-informed pre-training paradigm. To address these issues, this paper curates a large-scale protein-text paired dataset called ProtAnno with a property-driven sampling strategy, and introduces a novel function-informed protein pre-training paradigm. Specifically, the sampling strategy determines selecting probability based on the sample confidence and property coverage, balancing the data quality and data quantity in face of large-scale noisy data. Furthermore, motivated by significance of the protein specific functional mechanism, the proposed paradigm explicitly model protein static and dynamic functional segments by two segment-wise pre-training objectives, injecting fine-grained information in a function-informed manner. Leveraging all these innovations, we develop ProtCLIP, a multi-modality foundation model that comprehensively represents function-aware protein embeddings. On 22 different protein benchmarks within 5 types, including protein functionality classification, mutation effect prediction, cross-modal transformation, semantic similarity inference and protein-protein interaction prediction, our ProtCLIP consistently achieves SOTA performance, with remarkable improvements of 75% on average in five cross-modal transformation benchmarks, 59.9% in GO-CC and 39.7% in GO-BP protein function prediction. The experimental results verify the extraordinary potential of ProtCLIP serving as the protein multi-modality foundation model.
- Open Access
Bridge-IF: Learning Inverse Protein Folding with Markov Bridges
Inverse protein folding is a fundamental task in computational protein design, which aims to design protein sequences that fold into the desired backbone structures. While the development of machine learning algorithms for this task has seen significant success, the prevailing approaches, which predominantly employ a discriminative formulation, frequently encounter the error accumulation issue and often fail to capture the extensive variety of plausible sequences. To fill these gaps, we propose Bridge-IF, a generative diffusion bridge model for inverse folding, which is designed to learn the probabilistic dependency between the distributions of backbone structures and protein sequences. Specifically, we harness an expressive structure encoder to propose a discrete, informative prior derived from structures, and establish a Markov bridge to connect this prior with native sequences. During the inference stage, Bridge-IF progressively refines the prior sequence, culminating in a more plausible design. Moreover, we introduce a reparameterization perspective on Markov bridge models, from which we derive a simplified loss function that facilitates more effective training. We also modulate protein language models (PLMs) with structural conditions to precisely approximate the Markov bridge process, thereby significantly enhancing generation performance while maintaining parameter-efficient training. Extensive experiments on well-established benchmarks demonstrate that Bridge-IF predominantly surpasses existing baselines in sequence recovery and excels in the design of plausible proteins with high foldability. The code is available at https://github.com/violet-sto/Bridge-IF.
Open AccessBridge-IF: Learning Inverse Protein Folding with Markov Bridges
Inverse protein folding is a fundamental task in computational protein design, which aims to design protein sequences that fold into the desired backbone structures. While the development of machine learning algorithms for this task has seen significant success, the prevailing approaches, which predominantly employ a discriminative formulation, frequently encounter the error accumulation issue and often fail to capture the extensive variety of plausible sequences. To fill these gaps, we propose Bridge-IF, a generative diffusion bridge model for inverse folding, which is designed to learn the probabilistic dependency between the distributions of backbone structures and protein sequences. Specifically, we harness an expressive structure encoder to propose a discrete, informative prior derived from structures, and establish a Markov bridge to connect this prior with native sequences. During the inference stage, Bridge-IF progressively refines the prior sequence, culminating in a more plausible design. Moreover, we introduce a reparameterization perspective on Markov bridge models, from which we derive a simplified loss function that facilitates more effective training. We also modulate protein language models (PLMs) with structural conditions to precisely approximate the Markov bridge process, thereby significantly enhancing generation performance while maintaining parameter-efficient training. Extensive experiments on well-established benchmarks demonstrate that Bridge-IF predominantly surpasses existing baselines in sequence recovery and excels in the design of plausible proteins with high foldability. The code is available at https://github.com/violet-sto/Bridge-IF.
- Open Access
Preference or Intent? Double Disentangled Collaborative Filtering
People usually have different intents for choosing items, while their preferences under the same intent may also different. In traditional collaborative filtering approaches, both intent and preference factors are usually entangled in the modeling process, which significantly limits the robustness and interpretability of recommendation performances. For example, the low-rating items are always treated as negative feedback while they actually could provide positive information about user intent. To this end, in this paper, we propose a two-fold representation learning approach, namely Double Disentangled Collaborative Filtering (DDCF), for personalized recommendations. The first-level disentanglement is for separating the influence factors of intent and preference, while the second-level disentanglement is performed to build independent sparse preference representations under individual intent with limited computational complexity. Specifically, we employ two variational autoencoder networks, intent recognition network and preference decomposition network, to learn the intent and preference factors, respectively. In this way, the low-rating items will be treated as positive samples for modeling intents while the negative samples for modeling preferences. Finally, extensive experiments on three real-world datasets and four evaluation metrics clearly validate the effectiveness and the interpretability of DDCF.
Open AccessPreference or Intent? Double Disentangled Collaborative Filtering
People usually have different intents for choosing items, while their preferences under the same intent may also different. In traditional collaborative filtering approaches, both intent and preference factors are usually entangled in the modeling process, which significantly limits the robustness and interpretability of recommendation performances. For example, the low-rating items are always treated as negative feedback while they actually could provide positive information about user intent. To this end, in this paper, we propose a two-fold representation learning approach, namely Double Disentangled Collaborative Filtering (DDCF), for personalized recommendations. The first-level disentanglement is for separating the influence factors of intent and preference, while the second-level disentanglement is performed to build independent sparse preference representations under individual intent with limited computational complexity. Specifically, we employ two variational autoencoder networks, intent recognition network and preference decomposition network, to learn the intent and preference factors, respectively. In this way, the low-rating items will be treated as positive samples for modeling intents while the negative samples for modeling preferences. Finally, extensive experiments on three real-world datasets and four evaluation metrics clearly validate the effectiveness and the interpretability of DDCF.