Shuhan Qi

Jun Rao, Fei Wang, Liang Ding et al.

This article aims to provide the information retrieval community with some reflections on recent advances in retrieval learning by analyzing the reproducibility of image-text retrieval models. Due to the increase of multimodal data over the last decade, image-text retrieval has steadily become a major research direction in the field of information retrieval. Numerous researchers train and evaluate image-text retrieval algorithms using benchmark datasets such as MS-COCO and Flickr30k. Research in the past has mostly focused on performance, with multiple state-of-the-art methodologies being suggested in a variety of ways. According to their assertions, these techniques provide improved modality interactions and hence more precise multimodal representations. In contrast to previous works, we focus on the reproducibility of the approaches and the examination of the elements that lead to improved performance by pretrained and nonpretrained models in retrieving images and text. To be more specific, we first examine the related reproducibility concerns and explain why our focus is on image-text retrieval tasks. Second, we systematically summarize the current paradigm of image-text retrieval models and the stated contributions of those approaches. Third, we analyze various aspects of the reproduction of pretrained and nonpretrained retrieval models. To complete this, we conducted ablation experiments and obtained some influencing factors that affect retrieval recall more than the improvement claimed in the original paper. Finally, we present some reflections and challenges that the retrieval community should consider in the future. Our source code is publicly available at https://github.com/WangFei-2019/Image-text-Retrieval.

Jun Rao, Xv Meng, Liang Ding et al.

Knowledge distillation (KD) has been extensively employed to transfer the knowledge from a large teacher model to the smaller students, where the parameters of the teacher are fixed (or partially) during training. Recent studies show that this mode may cause difficulties in knowledge transfer due to the mismatched model capacities. To alleviate the mismatch problem, teacher-student joint training methods, e.g., online distillation, have been proposed, but it always requires expensive computational cost. In this paper, we present a parameter-efficient and student-friendly knowledge distillation method, namely PESF-KD, to achieve efficient and sufficient knowledge transfer by updating relatively few partial parameters. Technically, we first mathematically formulate the mismatch as the sharpness gap between their predictive distributions, where we show such a gap can be narrowed with the appropriate smoothness of the soft label. Then, we introduce an adapter module for the teacher and only update the adapter to obtain soft labels with appropriate smoothness. Experiments on a variety of benchmarks show that PESF-KD can significantly reduce the training cost while obtaining competitive results compared to advanced online distillation methods. Code will be released upon acceptance.

Jun Rao, Liang Ding, Shuhan Qi et al.

Although the vision-and-language pretraining (VLP) equipped cross-modal image-text retrieval (ITR) has achieved remarkable progress in the past two years, it suffers from a major drawback: the ever-increasing size of VLP models restricts its deployment to real-world search scenarios (where the high latency is unacceptable). To alleviate this problem, we present a novel plug-in dynamic contrastive distillation (DCD) framework to compress the large VLP models for the ITR task. Technically, we face the following two challenges: 1) the typical uni-modal metric learning approach is difficult to directly apply to the cross-modal tasks, due to the limited GPU memory to optimize too many negative samples during handling cross-modal fusion features. 2) it is inefficient to static optimize the student network from different hard samples, which have different effects on distillation learning and student network optimization. We try to overcome these challenges from two points. First, to achieve multi-modal contrastive learning, and balance the training costs and effects, we propose to use a teacher network to estimate the difficult samples for students, making the students absorb the powerful knowledge from pre-trained teachers, and master the knowledge from hard samples. Second, to dynamic learn from hard sample pairs, we propose dynamic distillation to dynamically learn samples of different difficulties, from the perspective of better balancing the difficulty of knowledge and students' self-learning ability. We successfully apply our proposed DCD strategy to two state-of-the-art vision-language pretrained models, i.e. ViLT and METER. Extensive experiments on MS-COCO and Flickr30K benchmarks show the effectiveness and efficiency of our DCD framework. Encouragingly, we can speed up the inference at least 129$\times$ compared to the existing ITR models.

Shaohuai Shi, Qing Yang, Yang Xiang et al.

To enable the pre-trained models to be fine-tuned with local data on edge devices without sharing data with the cloud, we design an efficient split fine-tuning (SFT) framework for edge and cloud collaborative learning. We propose three novel techniques in this framework. First, we propose a matrix decomposition-based method to compress the intermediate output of a neural network to reduce the communication volume between the edge device and the cloud server. Second, we eliminate particular links in the model without affecting the convergence performance in fine-tuning. Third, we implement our system atop PyTorch to allow users to easily extend their existing training scripts to enjoy the efficient edge and cloud collaborative learning. Experiments results on 9 NLP datasets show that our framework can reduce the communication traffic by 96 times with little impact on the model accuracy.

Dingkun Zhang, Shuhan Qi, Yulin Wu et al.

Multimodal Large Language Models (MLLMs) suffer from severe training inefficiency issue, which is associated with their massive model sizes and visual token numbers. Existing efforts in efficient training focus on reducing model sizes or trainable parameters. Inspired by the success of Visual Token Pruning (VTP) in improving inference efficiency, we are exploring another substantial research direction for efficient training by reducing visual tokens. However, applying VTP at the training stage results in a training-inference mismatch: pruning-trained models perform poorly when inferring on non-pruned full visual token sequences. To close this gap, we propose DualSpeed, a fast-slow framework for efficient training of MLLMs. The fast-mode is the primary mode, which incorporates existing VTP methods as plugins to reduce visual tokens, along with a mode isolator to isolate the model's behaviors. The slow-mode is the auxiliary mode, where the model is trained on full visual sequences to retain training-inference consistency. To boost its training, it further leverages self-distillation to learn from the sufficiently trained fast-mode. Together, DualSpeed can achieve both training efficiency and non-degraded performance. Experiments show DualSpeed accelerates the training of LLaVA-1.5 by 2.1$\times$ and LLaVA-NeXT by 4.0$\times$, retaining over 99% performance. Code: https://github.com/dingkun-zhang/DualSpeed

Shuhan Qi, Shuhao Zhang, Qiang Wang et al.

Value-decomposition methods, which reduce the difficulty of a multi-agent system by decomposing the joint state-action space into local observation-action spaces, have become popular in cooperative multi-agent reinforcement learning (MARL). However, value-decomposition methods still have the problems of tremendous sample consumption for training and lack of active exploration. In this paper, we propose a scalable value-decomposition exploration (SVDE) method, which includes a scalable training mechanism, intrinsic reward design, and explorative experience replay. The scalable training mechanism asynchronously decouples strategy learning with environmental interaction, so as to accelerate sample generation in a MapReduce manner. For the problem of lack of exploration, an intrinsic reward design and explorative experience replay are proposed, so as to enhance exploration to produce diverse samples and filter non-novel samples, respectively. Empirically, our method achieves the best performance on almost all maps compared to other popular algorithms in a set of StarCraft II micromanagement games. A data-efficiency experiment also shows the acceleration of SVDE for sample collection and policy convergence, and we demonstrate the effectiveness of factors in SVDE through a set of ablation experiments.

Shuhan Qi, Shuhao Zhang, Xiaohan Hou et al.

Multi-agent reinforcement learning for incomplete information environments has attracted extensive attention from researchers. However, due to the slow sample collection and poor sample exploration, there are still some problems in multi-agent reinforcement learning, such as unstable model iteration and low training efficiency. Moreover, most of the existing distributed framework are proposed for single-agent reinforcement learning and not suitable for multi-agent. In this paper, we design an distributed MARL framework based on the actor-work-learner architecture. In this framework, multiple asynchronous environment interaction modules can be deployed simultaneously, which greatly improves the sample collection speed and sample diversity. Meanwhile, to make full use of computing resources, we decouple the model iteration from environment interaction, and thus accelerate the policy iteration. Finally, we verified the effectiveness of propose framework in MaCA military simulation environment and the SMAC 3D realtime strategy gaming environment with imcomplete information characteristics.

Bohan Jin, Shuhan Qi, Kehai Chen et al.

The widespread use of Large Multimodal Models (LMMs) has raised concerns about model toxicity. However, current research mainly focuses on explicit toxicity, with less attention to some more implicit toxicity regarding prejudice and discrimination. To address this limitation, we introduce a subtler type of toxicity named dual-implicit toxicity and a novel toxicity benchmark termed MDIT-Bench: Multimodal Dual-Implicit Toxicity Benchmark. Specifically, we first create the MDIT-Dataset with dual-implicit toxicity using the proposed Multi-stage Human-in-loop In-context Generation method. Based on this dataset, we construct the MDIT-Bench, a benchmark for evaluating the sensitivity of models to dual-implicit toxicity, with 317,638 questions covering 12 categories, 23 subcategories, and 780 topics. MDIT-Bench includes three difficulty levels, and we propose a metric to measure the toxicity gap exhibited by the model across them. In the experiment, we conducted MDIT-Bench on 13 prominent LMMs, and the results show that these LMMs cannot handle dual-implicit toxicity effectively. The model's performance drops significantly in hard level, revealing that these LMMs still contain a significant amount of hidden but activatable toxicity. Data are available at https://github.com/nuo1nuo/MDIT-Bench.

Shuguang Jiao, Chengkai Huang, Shuhan Qi et al.

Agentic Retrieval-Augmented Generation (Agentic RAG) has become a widely adopted paradigm for multi-hop question answering and complex knowledge reasoning, where retrieval and reasoning are interleaved at inference time. As reasoning trajectories grow longer, failures become increasingly common. Existing approaches typically address such failures by either stopping at diagnostic analysis or rerunning the entire retrieval-reasoning pipeline, which leads to substantial computational overhead and redundant reasoning. In this paper, we propose Doctor-RAG (DR-RAG), a unified diagnose-and-repair framework that corrects failures in Agentic RAG through explicit error localization and prefix reuse, enabling minimal-cost intervention. DR-RAG decomposes failure handling into two consecutive stages: (i) trajectory-level failure diagnosis and localization, which attributes errors to a coverage-gated taxonomy and identifies the earliest failure point in the reasoning trajectory; and (ii) tool-conditioned local repair, which intervenes only at the diagnosed failure point while maximally reusing validated reasoning prefixes and retrieved evidence. By explicitly separating error attribution from correction, DR-RAG enables precise error localization, thereby avoiding expensive full-pipeline reruns and enabling targeted, efficient repair. We evaluate DR-RAG across three multi-hop question answering benchmarks, multiple agentic RAG baselines, and different backbone models. Experimental results demonstrate that DR-RAG substantially improves answer accuracy while significantly reducing reasoning token consumption compared to rerun-based repair strategies.

Hang Ren, Yulin Wu, Shuhan Qi et al.

Regret minimization is a powerful method for finding Nash equilibria in Normal-Form Games (NFGs) and Extensive-Form Games (EFGs), but it typically guarantees convergence only for the average strategy. However, computing the average strategy requires significant computational resources or introduces additional errors, limiting its practical applicability. The Reward Transformation (RT) framework was introduced to regret minimization to achieve last-iterate convergence through reward function regularization. However, it faces practical challenges: its performance is highly sensitive to manually tuned parameters, which often deviate from theoretical convergence conditions, leading to slow convergence, oscillations, or stagnation in local optima. Inspired by previous work, we propose an adaptive technique to address these issues, ensuring better consistency between theoretical guarantees and practical performance for RT Regret Matching (RTRM), RT Counterfactual Regret Minimization (RTCFR), and their variants in solving NFGs and EFGs more effectively. Our adaptive methods dynamically adjust parameters, balancing exploration and exploitation while improving regret accumulation, ultimately enhancing asymptotic last-iterate convergence and achieving linear convergence. Experimental results demonstrate that our methods significantly accelerate convergence, outperforming state-of-the-art algorithms.

Jiaxin Hong, Sixu Chen, Shuoyang Sun et al.

As 3D Gaussian Splatting (3DGS) emerges as a breakthrough in scene representation and novel view synthesis, its rapid adoption in safety-critical domains (e.g., autonomous systems, AR/VR) urgently demands scrutiny of potential security vulnerabilities. This paper presents the first systematic study of backdoor threats in 3DGS pipelines. We identify that adversaries may implant backdoor views to induce malicious scene confusion during inference, potentially leading to environmental misperception in autonomous navigation or spatial distortion in immersive environments. To uncover this risk, we propose GuassTrap, a novel poisoning attack method targeting 3DGS models. GuassTrap injects malicious views at specific attack viewpoints while preserving high-quality rendering in non-target views, ensuring minimal detectability and maximizing potential harm. Specifically, the proposed method consists of a three-stage pipeline (attack, stabilization, and normal training) to implant stealthy, viewpoint-consistent poisoned renderings in 3DGS, jointly optimizing attack efficacy and perceptual realism to expose security risks in 3D rendering. Extensive experiments on both synthetic and real-world datasets demonstrate that GuassTrap can effectively embed imperceptible yet harmful backdoor views while maintaining high-quality rendering in normal views, validating its robustness, adaptability, and practical applicability.

Dingkun Zhang, Shuhan Qi, Xinyu Xiao et al.

Recent advances in Multimodal Large Language Models (MLLMs) have enhanced their versatility as they integrate a growing number of modalities. Considering the heavy cost of training MLLMs, it is efficient to reuse the existing ones and extend them to more modalities through Modality-incremental Continual Learning (MCL). The exploration of MCL is in its early stages. In this work, we dive into the causes of performance degradation in MCL. We uncover that it suffers not only from forgetting as in traditional continual learning, but also from misalignment between the modality-agnostic and modality-specific components. To this end, we propose an elegantly simple MCL paradigm called "MErge then ReAlign" (MERA) to address both forgetting and misalignment. MERA avoids introducing heavy model budgets or modifying model architectures, hence is easy to deploy and highly reusable in the MLLM community. Extensive experiments demonstrate the impressive performance of MERA, holding an average of 99.84\% Backward Relative Gain when extending to four modalities, achieving nearly lossless MCL performance. Our findings underscore the misalignment issue in MCL. More broadly, our work showcases how to adjust different components of MLLMs during continual learning.

Jiahan Chen, Shuhan Qi, Yifan Li et al.

Databases are fundamental to contemporary information systems, yet traditional rule-based configuration methods struggle to manage the complexity of real-world applications with hundreds of tunable parameters. Deep reinforcement learning (DRL), which combines perception and decision-making, presents a potential solution for intelligent database configuration tuning. However, due to black-box property of RL-based method, the generated database tuning strategies still face the urgent problem of lack explainability. Besides, the redundant parameters in large scale database always make the strategy learning become unstable. This paper proposes KnobTree, an interpertable framework designed for the optimization of database parameter configuration. In this framework, an interpertable database tuning algorithm based on RL-based differentatial tree is proposed, which building a transparent tree-based model to generate explainable database tuning strategies. To address the problem of large-scale parameters, We also introduce a explainable method for parameter importance assessment, by utilizing Shapley Values to identify parameters that have significant impacts on database performance. Experiments conducted on MySQL and Gbase8s databases have verified exceptional transparency and interpretability of the KnobTree model. The good property makes generated strategies can offer practical guidance to algorithm designers and database administrators. Moreover, our approach also slightly outperforms the existing RL-based tuning algorithms in aspects such as throughput, latency, and processing time.

Xinhao Zhong, Hao Fang, Bin Chen et al.

Dataset distillation is an emerging dataset reduction method, which condenses large-scale datasets while maintaining task accuracy. Current parameterization methods achieve enhanced performance under extremely high compression ratio by optimizing determined synthetic dataset in informative feature domain. However, they limit themselves to a fixed optimization space for distillation, neglecting the diverse guidance across different informative latent spaces. To overcome this limitation, we propose a novel parameterization method dubbed Hierarchical Parameterization Distillation (H-PD), to systematically explore hierarchical feature within provided feature space (e.g., layers within pre-trained generative adversarial networks). We verify the correctness of our insights by applying the hierarchical optimization strategy on GAN-based parameterization method. In addition, we introduce a novel class-relevant feature distance metric to alleviate the computational burden associated with synthetic dataset evaluation, bridging the gap between synthetic and original datasets. Experimental results demonstrate that the proposed H-PD achieves a significant performance improvement under various settings with equivalent time consumption, and even surpasses current generative distillation using diffusion models under extreme compression ratios IPC=1 and IPC=10.

Huale Li, Xuan Wang, Zengyue Guo et al.

Counterfactual Regret Minimization (CFR)} is the popular method for finding approximate Nash equilibrium in two-player zero-sum games with imperfect information. CFR solves games by travsersing the full game tree iteratively, which limits its scalability in larger games. When applying CFR to solve large-scale games in previously, large-scale games are abstracted into small-scale games firstly. Secondly, CFR is used to solve the abstract game. And finally, the solution strategy is mapped back to the original large-scale game. However, this process requires considerable expert knowledge, and the accuracy of abstraction is closely related to expert knowledge. In addition, the abstraction also loses certain information, which will eventually affect the accuracy of the solution strategy. Towards this problem, a recent method, \textit{Deep CFR} alleviates the need for abstraction and expert knowledge by applying deep neural networks directly to CFR in full games. In this paper, we introduces \textit{Neural Network Counterfactual Regret Minimization (NNCFR)}, an improved variant of \textit{Deep CFR} that has a faster convergence by constructing a dueling netwok as the value network. Moreover, an evaluation module is designed by combining the value network and Monte Carlo, which reduces the approximation error of the value network. In addition, a new loss function is designed in the procedure of training policy network in the proposed \textit{NNCFR}, which can be good to make the policy network more stable. The extensive experimental tests are conducted to show that the \textit{NNCFR} converges faster and performs more stable than \textit{Deep CFR}, and outperforms \textit{Deep CFR} with respect to exploitability and head-to-head performance on test games.

Huale Li, Xuan Wang, Fengwei Jia et al.

Counterfactual regret minimization (CFR) is a popular method to deal with decision-making problems of two-player zero-sum games with imperfect information. Unlike existing studies that mostly explore for solving larger scale problems or accelerating solution efficiency, we propose a framework, RLCFR, which aims at improving the generalization ability of the CFR method. In the RLCFR, the game strategy is solved by the CFR in a reinforcement learning framework. And the dynamic procedure of iterative interactive strategy updating is modeled as a Markov decision process (MDP). Our method, RLCFR, then learns a policy to select the appropriate way of regret updating in the process of iteration. In addition, a stepwise reward function is formulated to learn the action policy, which is proportional to how well the iteration strategy is at each step. Extensive experimental results on various games have shown that the generalization ability of our method is significantly improved compared with existing state-of-the-art methods.