Publications on Minghui Chen

Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization

Mon, 28 Sep 2026 00:00:00 +0000

Authors: Xuanqi Zhang, Ruinan Jin, Running Yang, Yuxuan Zhang, Minghui Chen, Wenlong Deng, Xiaoxiao Li

Publication: arXiv preprint, 2026

Overview

Likelihood-Only Tool Scoring (LOTS) measures how removing an observed tool output changes the likelihood of an agent’s answer. It accumulates these measurements across recurring tasks to maintain a useful tool set. Experiments examine improvements over time, reductions in tool context, and transfer of learned tool selections between models.

For-Value: Efficient Forward-Only Data Valuation for finetuning LLMs and VLMs

Wed, 01 Jul 2026 00:00:00 +0000

Authors: Wenlong Deng, Qi Zeng, Jiaming Zhang, Minghui Chen, Zixin Ding, Christos Thrampoulidis, Boying Gong, Xiaoxiao Li

Published in: Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics, Volume 1: Long Papers (ACL 2026), July 2026, pp. 14581–14600

Overview

For-Value estimates data influence using final-layer representations and prediction errors. A closed-form calculation replaces backpropagation, allowing examples to be evaluated in batches with a single forward pass. The paper evaluates the method on influential-data detection and mislabeled-data identification for language and vision-language models.

LatentGym: A Testbed For Cross-Task Experiential Learning With Controllable Latent Structure

Sat, 13 Jun 2026 00:00:00 +0000

Authors: Daksh Mittal, Tommaso Castellani, Thomson Yen, Naimeng Ye, Fangyu Wu, Minghui Chen, Tiffany Cai, Emmanouil Koukoumidis, William Zeng, Hongseok Namkoong

Publication: arXiv preprint, 2026

Overview

LatentGym provides task sequences governed by shared, controllable hidden variables. Its evaluation separates an agent’s ability to discover that structure from its ability to use the information in later decisions. The study examines adaptation failures, post-training on related tasks, and the effects of feedback between tasks.

Forgettable Federated Linear Learning with Certified Data Unlearning

Sun, 24 May 2026 00:00:00 +0000

Authors: Ruinan Jin, Minghui Chen, Qiong Zhang, Xiaoxiao Li

Published in: IEEE Transactions on Neural Networks and Learning Systems (IEEE TNNLS, 2026), Early Access, pp. 1–10

Earlier workshop version: Forgettable Federated Linear Learning with Certified Data Removal, WWW 2024 FL@FM Workshop (Best Paper Award 🏆)

Overview

Forgettable Federated Linear Learning uses pre-trained models to approximate deep networks linearly during federated training. Its certified unlearning procedure lets the server remove a target client’s contribution without contacting clients again or retaining historical models. Experiments spanning small and large datasets, convolutional networks, and foundation models evaluate the balance between model accuracy and effective unlearning.

Textual Equilibrium Propagation for Deep Compound AI Systems

Wed, 28 Jan 2026 00:00:00 +0000

Authors: Minghui Chen, Wenlong Deng, James Zou, Han Yu, Xiaoxiao Li

Published in: Accepted to The Fourteenth International Conference on Learning Representations (ICLR 2026)

Abstract

Large language models (LLMs) are increasingly deployed as part of compound AI systems that coordinate multiple modules, such as retrievers, tools, and verifiers, over long-horizon workflows. Recent approaches that propagate textual feedback globally, such as TextGrad, make it feasible to optimize such pipelines, but we find that performance degrades as system depth grows.

A Closer Look at Personalized Fine-Tuning in Heterogeneous Federated Learning

Sun, 16 Nov 2025 00:00:00 +0000

Authors: Minghui Chen, Hrad Ghoukasian, Ruinan Jin, Zehua Wang, Sai Praneeth Karimireddy, Xiaoxiao Li

Publication: arXiv preprint, 2025

Overview

Local fine-tuning can distort features learned by a federated global model, reducing generalization under client heterogeneity. This work studies linear probing followed by full fine-tuning across seven datasets and six personalization variants. Its empirical and theoretical analyses identify conditions in which phased updates provide a better balance between local adaptation and generalization.

Enhancing Clinical Multiple-Choice Questions Benchmarks with Knowledge Graph Guided Distractor Generation

Sat, 31 May 2025 00:00:00 +0000

Authors: Running Yang (Equal contribution), Wenlong Deng (Equal contribution), Minghui Chen, Yuyin Zhou, Xiaoxiao Li

Published in: ICML 2025 Workshop on Reliable and Responsible Foundation Models

Overview

Knowledge Graph Guided Distractor Generation (KGGDG) follows medical knowledge-graph paths to construct plausible but incorrect answers to clinical multiple-choice questions. Evaluations on six medical QA benchmarks show that these answer choices expose weaknesses that easier distractors can conceal, supporting more demanding assessments of medical LLMs.

Can Textual Gradient Work in Federated Learning?

Fri, 24 Jan 2025 00:00:00 +0000

Authors: Minghui Chen, Ruinan Jin, Wenlong Deng, Yuanyuan Chen, Zhi Huang, Han Yu, Xiaoxiao Li

Published in: The Thirteenth International Conference on Learning Representations (ICLR 2025)

Abstract

Recent studies highlight the promise of LLM-based prompt optimization, especially with TextGrad, which automates “differentiation” via texts and backpropagates textual feedback provided by LLMs. This approach facilitates training in various real-world applications that do not support numerical gradient propagation or loss calculation. It opens new avenues for optimization in decentralized, resource-constrained environments, suggesting that users of black-box LLMs (e.g., ChatGPT) could enhance components of LLM agentic systems (such as prompt optimization) through collaborative paradigms like federated learning (FL).

DARE the Extreme: Revisiting Delta-Parameter Pruning For Fine-Tuned Models

Thu, 23 Jan 2025 00:00:00 +0000

Authors: Wenlong Deng, Yize Zhao, Vala Vakilian, Minghui Chen, Xiaoxiao Li, Christos Thrampoulidis

Published in: The Thirteenth International Conference on Learning Representations (ICLR 2025, Spotlight)

Abstract

Storing open-source fine-tuned models separately introduces redundancy and increases response times in applications utilizing multiple models. Delta-parameter pruning (DPP), particularly the random drop and rescale (DARE) method proposed by Yu et al., addresses this by pruning the majority of delta parameters—the differences between fine-tuned and pre-trained model weights—while typically maintaining minimal performance loss. However, DARE fails when either the pruning rate or the magnitude of the delta parameters is large. We highlight two key reasons for this failure - (1) an excessively large rescaling factor as pruning rates increase, and (2) high mean and variance in the delta parameters. To address these, we develop two algorithmic improvements - (1) DARq, which modifies the rescaling factor in DARE, leading to significant performance gains at high pruning rates (e.g., >30% on COLA and SST2 for encoder models, with even larger improvements in decoder models), and (2) AdamR, an in-training modification that incorporates appropriate Delta regularization before applying DPP. We also demonstrate that DARq can be seamlessly combined with vanilla parameter-efficient fine-tuning techniques like LoRA and can facilitate structural DPP. Additionally, we revisit the application of importance-based pruning techniques within DPP, demonstrating that they outperform random-based methods when delta parameters are large. Through this comprehensive study, we develop a pipeline for selecting the most appropriate DPP method under various practical scenarios.

Local Superior Soups: A Catalyst for Model Merging in Cross-Silo Federated Learning

Tue, 15 Oct 2024 00:00:00 +0000

Authors: Minghui Chen, Meirui Jiang, Xin Zhang, Qi Dou, Zehua Wang, Xiaoxiao Li

Published in: The Thirty-Eighth Conference on Neural Information Processing Systems (NeurIPS 2024)

Abstract

Federated learning (FL) is a learning paradigm that enables collaborative training of models using decentralized data. Recently, the utilization of pre-trained weight initialization in FL has been demonstrated to effectively improve model performance. However, the evolving complexity of current pre-trained models, characterized by a substantial increase in parameters, markedly intensifies the challenges associated with communication rounds required for their adaptation to FL. To address these communication cost issues and increase the performance of pre-trained model adaptation in FL, we propose an innovative model interpolation-based local training technique called “Local Superior Soups.” Our method enhances local training across different clients, encouraging the exploration of a connected low-loss basin within a few communication rounds through regularized model interpolation. This approach acts as a catalyst for the seamless adaptation of pre-trained models in in FL. We demonstrated its effectiveness and efficiency across diverse widely-used FL datasets.

Debiased Noise Editing on Foundation Models for Fair Medical Image Classification

Fri, 20 Sep 2024 00:00:00 +0000

Authors: Ruinan Jin, Wenlong Deng, Minghui Chen, Xiaoxiao Li

Published in: The 27th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2024)

Abstract

In the era of Foundation Models’ (FMs) rising prominence in AI, our study addresses the challenge of biases in medical images while the model operates in black-box (e.g., using FM API), particularly spurious correlations between pixels and sensitive attributes. Traditional methods for bias mitigation face limitations due to the restricted access to web-hosted FMs and difficulties in addressing the underlying bias encoded within the FM API. We propose a D(ebiased) N(oise) E(diting) strategy, termed DNE, which generates DNE noise to mask such spurious correlation. DNE is capable of mitigating bias both within the FM API embedding and the images themselves. Furthermore, DNE is suitable for both white-box and black-box FM APIs, where we introduced G(reedy) (Z)eroth-O(rder) (GeZO) optimization for it when the gradient is inaccessible in black-box APIs. Our whole pipeline enables fairness-aware image editing that can be applied across various medical contexts without requiring direct model manipulation or significant computational resources. Our empirical results demonstrate the method’s effectiveness in maintaining fairness and utility across different patient groups and diseases. In the era of AI-driven medicine, this work contributes to making healthcare diagnostics more equitable, showcasing a practical solution for bias mitigation in pre-trained image FMs.

FedSoup: Improving Generalization and Personalization in Federated Learning via Selective Model Interpolation

Thu, 20 Jul 2023 00:00:00 +0000

Authors: Minghui Chen, Meirui Jiang, Qi Dou, Zehua Wang, Xiaoxiao Li

Published in: The 26th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2023)

Abstract

Cross-silo federated learning (FL) enables the development of machine learning models on datasets distributed across data centers such as hospitals and clinical research laboratories. However, recent research has found that current FL algorithms face a trade-off between local and global performance when confronted with distribution shifts. Specifically, personalized FL methods have a tendency to overfit to local data, leading to a sharp valley in the local model and inhibiting its ability to generalize to out-of-distribution data. In this paper, we propose a novel federated model soup method (i.e., selective interpolation of model parameters) to optimize the trade-off between local and global performance. Specifically, during the federated training phase, each client maintains its own global model pool by monitoring the performance of the interpolated model between the local and global models. This allows us to alleviate overfitting and seek flat minima, which can significantly improve the model’s generalization performance.

VITA: A Multi-Source Vicinal Transfer Augmentation Method for Out-of-Distribution Generalization

Tue, 01 Feb 2022 00:00:00 +0000

Authors: Minghui Chen, Cheng Wen, Feng Zheng, Fengxiang He, Ling Shao

Published in: Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2022)

Abstract

Invariance to diverse types of image corruption, such as noise, blurring, or colour shifts, is essential to establish robust models in computer vision. Data augmentation has been the major approach in improving the robustness against common corruptions. However, the samples produced by popular augmentation strategies deviate significantly from the underlying data manifold. As a result, performance is skewed toward certain types of corruption. To address this issue, we propose a multi-source vicinal transfer augmentation (VITA) method for generating diverse on-manifold samples. The proposed VITA consists of two complementary parts, tangent transfer and integration of multi-source vicinal samples. The tangent transfer creates initial augmented samples for improving corruption robustness. The integration employs a generative model to characterize the underlying manifold built by vicinal samples, facilitating the generation of on-manifold samples. Our proposed VITA significantly outperforms the current state-of-the-art augmentation methods, demonstrated in extensive experiments on corruption benchmarks.

Benchmarks for Corruption Invariant Person Re-Identification

Wed, 01 Dec 2021 00:00:00 +0000

Authors: Minghui Chen (Equal contribution), Zhiqiang Wang (Equal contribution), Feng Zheng

Published in: Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS 2021)

Abstract

When deploying person re-identification (ReID) model in safety-critical applications, it is pivotal to understanding the robustness of the model against a diverse array of image corruptions. However, current evaluations of person ReID only consider the performance on clean datasets and ignore images in various corrupted scenarios. In this work, we comprehensively establish five ReID benchmarks for learning corruption invariant representation. In the field of ReID, we are the first to conduct an exhaustive study on corruption invariant learning in single- and cross-modality datasets, including Market-1501, CUHK03, MSMT17, RegDB, SYSU-MM01. After reproducing and examining the robustness performance of 21 recent ReID methods, we have some observations, 1) transformer-based models are more robust towards corrupted images, compared with CNN-based models, 2) increasing the probability of random erasing (a commonly used augmentation method) hurts model corruption robustness, 3) cross-dataset generalization improves with corruption robustness increases. By analyzing the above observations, we propose a strong baseline on both single- and cross-modality ReID datasets which achieves improved robustness against diverse corruptions. Our codes are available on github( https://github.com/MinghuiChen43/CIL-ReID).