arXiv subjects

Yancheng Yuan

Publications and source records attributed to Yancheng Yuan.

At least 37 records · Page 2Linked to original sources

arXiv2026

MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering

Multimodal large language models (MLLMs) have demonstrated remarkable capabilities across vision-language tasks, yet their large-scale deployment raises pressing concerns about memorized private data, outdated knowledge, and harmful content. Existing unlearning approaches for MLLMs typically adapt training-based strategies such as gradient ascent or preference optimization, but these methods are computationally expensive, irreversible, and often distort retained knowledge. In this work, we propose MLLMEraser, an input-aware, training-free framework for test-time unlearning. Our approach leverages activation steering to enable dynamic knowledge erasure without parameter updates. Specifically, we construct a multimodal erasure direction by contrasting adversarially perturbed, knowledge-recall image-text pairs with knowledge-erasure counterparts, capturing both textual and visual discrepancies. To prevent unnecessary interference, we further design an input-aware steering mechanism that adaptively determines when and how the erasure direction should be applied, preserving utility on retained knowledge while enforcing forgetting on designated content. Experiments on LLaVA-1.5 and Qwen-2.5-VL demonstrate that MLLMEraser consistently outperforms state-of-the-art MLLM unlearning baselines, achieving stronger forgetting performance with lower computational cost and minimal utility degradation.

cs.LG↗

arXiv2026

Think Dense, Not Long: Dynamic Decoupled Conditional Advantage for Efficient Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) can elicit strong multi-step reasoning, yet it often encourages overly verbose traces. Moreover, naive length penalties in group-relative optimization can severely hurt accuracy. We attribute this failure to two structural issues: (i) Dilution of Length Baseline, where incorrect responses (with zero length reward) depress the group baseline and over-penalize correct solutions; and (ii) Difficulty-Penalty Mismatch, where a static penalty cannot adapt to problem difficulty, suppressing necessary reasoning on hard instances while leaving redundancy on easy ones. We propose Dynamic Decoupled Conditional Advantage (DDCA) to decouple efficiency optimization from correctness. DDCA computes length advantages conditionally within the correct-response cluster to eliminate baseline dilution, and dynamically scales the penalty strength using the group pass rate as a proxy for difficulty. Experiments on GSM8K, MATH500, AMC23, and AIME25 show that DDCA consistently improves the efficiency--accuracy trade-off relative to adaptive baselines, reducing generated tokens by approximately 60% on simpler tasks (e.g., GSM8K) versus over 20% on harder benchmarks (e.g., AIME25), thereby maintaining or improving accuracy. Code is available at https://github.com/alphadl/DDCA.

cs.CL↗

arXiv2025

Magnitude-Modulated Equivariant Adapter for Parameter-Efficient Fine-Tuning of Equivariant Graph Neural Networks

Pretrained equivariant graph neural networks based on spherical harmonics offer efficient and accurate alternatives to computationally expensive ab-initio methods, yet adapting them to new tasks and chemical environments still requires fine-tuning. Conventional parameter-efficient fine-tuning (PEFT) techniques, such as Adapters and LoRA, typically break symmetry, making them incompatible with those equivariant architectures. ELoRA, recently proposed, is the first equivariant PEFT method. It achieves improved parameter efficiency and performance on many benchmarks. However, the relatively high degrees of freedom it retains within each tensor order can still perturb pretrained feature distributions and ultimately degrade performance. To address this, we present Magnitude-Modulated Equivariant Adapter (MMEA), a novel equivariant fine-tuning method which employs lightweight scalar gating to modulate feature magnitudes on a per-order and per-multiplicity basis. We demonstrate that MMEA preserves strict equivariance and, across multiple benchmarks, consistently improves energy and force predictions to state-of-the-art levels while training fewer parameters than competing approaches. These results suggest that, in many practical scenarios, modulating channel magnitudes is sufficient to adapt equivariant models to new chemical environments without breaking symmetry, pointing toward a new paradigm for equivariant PEFT design.

cs.LG↗

arXiv2025

On the Convergence Analysis of an Inexact Preconditioned Stochastic Model-Based Algorithm

This paper focuses on investigating an inexact stochastic model-based optimization algorithm that integrates preconditioning techniques for solving stochastic composite optimization problems. The proposed framework unifies and extends the fixed-metric stochastic model-based algorithm to its preconditioned and inexact variants. Convergence guarantees are established under mild assumptions for both weakly convex and convex settings, without requiring smoothness or global Lipschitz continuity of the objective function. By assuming a local Lipschitz condition, we derive nonasymptotic and asymptotic convergence rates measured by the gradient of the Moreau envelope. Furthermore, convergence rates in terms of the distance to the optimal solution set are obtained under an additional quadratic growth condition on the objective function. Numerical experiment results demonstrate the theoretical findings for the proposed algorithm.

math.OC↗

arXiv2025

dHPR: A Distributed Halpern Peaceman--Rachford Method for Non-smooth Distributed Optimization Problems

This paper introduces the distributed Halpern Peaceman--Rachford (dHPR) method, an efficient algorithm for solving distributed convex composite optimization problems with non-smooth objectives, which achieves a non-ergodic $O(1/k)$ iteration complexity regarding Karush--Kuhn--Tucker residual. By leveraging the symmetric Gauss--Seidel decomposition, the dHPR effectively decouples the linear operators in the objective functions and consensus constraints while maintaining parallelizability and avoiding additional large proximal terms, leading to a decentralized implementation with provably fast convergence. The superior performance of dHPR is demonstrated through comprehensive numerical experiments on distributed LASSO, group LASSO, and $L_1$-regularized logistic regression problems.

math.OC↗

arXiv2025

On the Relationships among GPU-Accelerated First-Order Methods for Solving Linear Programming

This paper aims to understand the relationships among recently developed GPU-accelerated first-order methods (FOMs) for linear programming (LP), with particular emphasis on HPR-LP -- a Halpern Peaceman--Rachford (HPR) method for LP. Our findings can be summarized as follows: (i) the base algorithm of cuPDLPx, a recently released GPU solver, is a special case of the base algorithm of HPR-LP, thereby showing that cuPDLPx is another concrete implementation instance of HPR-LP; (ii) once the active sets have been identified, HPR-LP and EPR-LP -- an ergodic PR method for LP -- become equivalent under the same initialization; and (iii) extensive numerical experiments on benchmark datasets demonstrate that HPR-LP achieves the best overall performance among current GPU-accelerated LP solvers. These findings provide a strong motivation for using the HPR method as a baseline to further develop GPU-accelerated LP solvers and beyond.

math.OC↗

arXiv2025

A Survey on Large Language Model-based Agents for Statistics and Data Science

In recent years, data science agents powered by Large Language Models (LLMs), known as "data agents," have shown significant potential to transform the traditional data analysis paradigm. This survey provides an overview of the evolution, capabilities, and applications of LLM-based data agents, highlighting their role in simplifying complex data tasks and lowering the entry barrier for users without related expertise. We explore current trends in the design of LLM-based frameworks, detailing essential features such as planning, reasoning, reflection, multi-agent collaboration, user interface, knowledge integration, and system design, which enable agents to address data-centric problems with minimal human intervention. Furthermore, we analyze several case studies to demonstrate the practical applications of various data agents in real-world scenarios. Finally, we identify key challenges and propose future research directions to advance the development of data agents into intelligent statistical analysis software.

cs.AI↗

arXiv2025

Delayed Feedback Modeling with Influence Functions

In online advertising under the cost-per-conversion (CPA) model, accurate conversion rate (CVR) prediction is crucial. A major challenge is delayed feedback, where conversions may occur long after user interactions, leading to incomplete recent data and biased model training. Existing solutions partially mitigate this issue but often rely on auxiliary models, making them computationally inefficient and less adaptive to user interest shifts. We propose IF-DFM, an \underline{I}nfluence \underline{F}unction-empowered for \underline{D}elayed \underline{F}eedback \underline{M}odeling which estimates the impact of newly arrived and delayed conversions on model parameters, enabling efficient updates without full retraining. By reformulating the inverse Hessian-vector product as an optimization problem, IF-DFM achieves a favorable trade-off between scalability and effectiveness. Experiments on benchmark datasets show that IF-DFM outperforms prior methods in both accuracy and adaptability.

cs.LG↗

arXiv2025

HPR-QP: A dual Halpern Peaceman-Rachford method for solving large-scale convex composite quadratic programming

In this paper, we introduce HPR-QP, a dual Halpern Peaceman-Rachford (HPR) method designed for solving large-scale convex composite quadratic programming. One distinctive feature of HPR-QP is that, instead of working with the primal formulations, it builds on the novel restricted Wolfe dual introduced in recent years. It also leverages the symmetric Gauss-Seidel technique to simplify subproblem updates without introducing auxiliary slack variables that typically lead to slow convergence. By restricting updates to the range space of the Hessian of the quadratic objective function, HPR-QP employs proximal operators of smaller spectral norms to speed up the convergence. Shadow sequences are elaborately constructed to deal with the range space constraints. Additionally, HPR-QP incorporates adaptive restart and penalty parameter update strategies, derived from the HPR method's $O(1/k)$ convergence in terms of the Karush-Kuhn-Tucker residual, to further enhance its performance and robustness. Extensive numerical experiments on benchmark data sets using a GPU demonstrate that our Julia implementation of HPR-QP significantly outperforms state-of-the-art solvers in both speed and scalability.

math.OC↗

arXiv2025

Accelerating RLHF Training with Reward Variance Increase

Reinforcement learning from human feedback (RLHF) is an essential technique for ensuring that large language models (LLMs) are aligned with human values and preferences during the post-training phase. As an effective RLHF approach, group relative policy optimization (GRPO) has demonstrated success in many LLM-based applications. However, efficient GRPO-based RLHF training remains a challenge. Recent studies reveal that a higher reward variance of the initial policy model leads to faster RLHF training. Inspired by this finding, we propose a practical reward adjustment model to accelerate RLHF training by provably increasing the reward variance and preserving the relative preferences and reward expectation. Our reward adjustment method inherently poses a nonconvex optimization problem, which is NP-hard to solve in general. To overcome the computational challenges, we design a novel $O(n \log n)$ algorithm to find a global solution of the nonconvex reward adjustment model by explicitly characterizing the extreme points of the feasible set. As an important application, we naturally integrate this reward adjustment model into the GRPO algorithm, leading to a more efficient GRPO with reward variance increase (GRPOVI) algorithm for RLHF training. As an interesting byproduct, we provide an indirect explanation for the empirical effectiveness of GRPO with rule-based reward for RLHF training, as demonstrated in DeepSeek-R1. Experiment results demonstrate that the GRPOVI algorithm can significantly improve the RLHF training efficiency compared to the original GRPO algorithm.

cs.LG↗

arXiv2025

Enhancing Input-Label Mapping in In-Context Learning with Contrastive Decoding

Large language models (LLMs) excel at a range of tasks through in-context learning (ICL), where only a few task examples guide their predictions. However, prior research highlights that LLMs often overlook input-label mapping information in ICL, relying more on their pre-trained knowledge. To address this issue, we introduce In-Context Contrastive Decoding (ICCD), a novel method that emphasizes input-label mapping by contrasting the output distributions between positive and negative in-context examples. Experiments on 7 natural language understanding (NLU) tasks show that our ICCD method brings consistent and significant improvement (up to +1.8 improvement on average) upon 6 different scales of LLMs without requiring additional training. Our approach is versatile, enhancing performance with various demonstration selection methods, demonstrating its broad applicability and effectiveness. The code and scripts are released at https://github.com/Romainpkq/CD_ICL.

cs.CL↗

arXiv2025

LAMBDA: A Large Model Based Data Agent

We introduce LArge Model Based Data Agent (LAMBDA), a novel open-source, code-free multi-agent data analysis system that leverages the power of large language models. LAMBDA is designed to address data analysis challenges in data-driven applications through innovatively designed data agents using natural language. At the core of LAMBDA are two key agent roles: the programmer and the inspector, which are engineered to work together seamlessly. Specifically, the programmer generates code based on the user's instructions and domain-specific knowledge, while the inspector debugs the code when necessary. To ensure robustness and handle adverse scenarios, LAMBDA features a user interface that allows direct user intervention. Moreover, LAMBDA can flexibly integrate external models and algorithms through our proposed Knowledge Integration Mechanism, catering to the needs of customized data analysis. LAMBDA has demonstrated strong performance on various data analysis tasks. It has the potential to enhance data analysis paradigms by seamlessly integrating human and artificial intelligence, making it more accessible, effective, and efficient for users from diverse backgrounds. The strong performance of LAMBDA in solving data analysis problems is demonstrated using real-world data examples. The code for LAMBDA is available at https://github.com/AMA-CMFAI/LAMBDA and videos of three case studies can be viewed at https://www.polyu.edu.hk/ama/cmfai/lambda.html.

cs.AI↗

arXiv2025

Addressing Missing Data Issue for Diffusion-based Recommendation

Diffusion models have shown significant potential in generating oracle items that best match user preference with guidance from user historical interaction sequences. However, the quality of guidance is often compromised by unpredictable missing data in observed sequence, leading to suboptimal item generation. Since missing data is uncertain in both occurrence and content, recovering it is impractical and may introduce additional errors. To tackle this challenge, we propose a novel dual-side Thompson sampling-based Diffusion Model (TDM), which simulates extra missing data in the guidance signals and allows diffusion models to handle existing missing data through extrapolation. To preserve user preference evolution in sequences despite extra missing data, we introduce Dual-side Thompson Sampling to implement simulation with two probability models, sampling by exploiting user preference from both item continuity and sequence stability. TDM strategically removes items from sequences based on dual-side Thompson sampling and treats these edited sequences as guidance for diffusion models, enhancing models' robustness to missing data through consistency regularization. Additionally, to enhance the generation efficiency, TDM is implemented under the denoising diffusion implicit models to accelerate the reverse process. Extensive experiments and theoretical analysis validate the effectiveness of TDM in addressing missing data in sequential recommendations.

cs.IR↗

arXiv2025

Adaptive sieving: A dimension reduction technique for sparse optimization problems

In this paper, we propose an adaptive sieving (AS) strategy for solving general sparse machine learning models by effectively exploring the intrinsic sparsity of the solutions, wherein only a sequence of reduced problems with much smaller sizes need to be solved. We further apply the proposed AS strategy to generate solution paths for large-scale sparse optimization problems efficiently. We establish the theoretical guarantees for the proposed AS strategy including its finite termination property. Extensive numerical experiments are presented in this paper to demonstrate the effectiveness and flexibility of the AS strategy to solve large-scale machine learning models.

math.OC↗

arXiv2025

Unified Parameter-Efficient Unlearning for LLMs

The advent of Large Language Models (LLMs) has revolutionized natural language processing, enabling advanced understanding and reasoning capabilities across a variety of tasks. Fine-tuning these models for specific domains, particularly through Parameter-Efficient Fine-Tuning (PEFT) strategies like LoRA, has become a prevalent practice due to its efficiency. However, this raises significant privacy and security concerns, as models may inadvertently retain and disseminate sensitive or undesirable information. To address these issues, we introduce a novel instance-wise unlearning framework, LLMEraser, which systematically categorizes unlearning tasks and applies precise parameter adjustments using influence functions. Unlike traditional unlearning techniques that are often limited in scope and require extensive retraining, LLMEraser is designed to handle a broad spectrum of unlearning tasks without compromising model performance. Extensive experiments on benchmark datasets demonstrate that LLMEraser excels in efficiently managing various unlearning scenarios while maintaining the overall integrity and efficacy of the models.

cs.AI↗

arXiv2025

HOT: An Efficient Halpern Accelerating Algorithm for Optimal Transport Problems

This paper proposes an efficient HOT algorithm for solving the optimal transport (OT) problems with finite supports. We particularly focus on an efficient implementation of the HOT algorithm for the case where the supports are in $\mathbb{R}^2$ with ground distances calculated by $L_2^2$-norm. Specifically, we design a Halpern accelerating algorithm to solve the equivalent reduced model of the discrete OT problem. Moreover, we derive a novel procedure to solve the involved linear systems in the HOT algorithm in linear time complexity. Consequently, we can obtain an $\varepsilon$-approximate solution to the optimal transport problem with $M$ supports in $O(M^{1.5}/\varepsilon)$ flops, which significantly improves the best-known computational complexity. We further propose an efficient procedure to recover an optimal transport plan for the original OT problem based on a solution to the reduced model, thereby overcoming the limitations of the reduced OT model in applications that require the transport plan. We implement the HOT algorithm in PyTorch and extensive numerical results show the superior performance of the HOT algorithm compared to existing state-of-the-art algorithms for solving the OT problems.

math.OC↗

arXiv2025

HPR-LP: An implementation of an HPR method for solving linear programming

In this paper, we introduce an HPR-LP solver, an implementation of a Halpern Peaceman-Rachford (HPR) method with semi-proximal terms for solving linear programming (LP). The HPR method enjoys the iteration complexity of $O(1/k)$ in terms of the Karush-Kuhn-Tucker residual and the objective error. Based on the complexity results, we design an adaptive strategy of restart and penalty parameter update to improve the efficiency and robustness of the HPR method. We conduct extensive numerical experiments on different LP benchmark datasets using NVIDIA A100-SXM4-80GB GPU in different stopping tolerances. Our solver's Julia version achieves a $\textbf{2.39x}$ to $\textbf{5.70x}$ speedup measured by SGM10 on benchmark datasets with presolve ($\textbf{2.03x}$ to $\textbf{4.06x}$ without presolve) over the award-winning solver PDLP with the tolerance of $10^{-8}$.

math.OC↗

arXiv2025

PyClustrPath: An efficient Python package for generating clustering paths with GPU acceleration

Convex clustering is a popular clustering model without requiring the number of clusters as prior knowledge. It can generate a clustering path by continuously solving the model with a sequence of regularization parameter values. This paper introduces {\it PyClustrPath}, a highly efficient Python package for solving the convex clustering model with GPU acceleration. {\it PyClustrPath} implements popular first-order and second-order algorithms with a clean modular design. Such a design makes {\it PyClustrPath} more scalable to incorporate new algorithms for solving the convex clustering model in the future. We extensively test the numerical performance of {\it PyClustrPath} on popular clustering datasets, demonstrating its superior performance compared to the existing solvers for generating the clustering path based on the convex clustering model. The implementation of {\it PyClustrPath} can be found at: https://github.com/D3IntOpt/PyClustrPath.

math.OC↗