arXiv subjects

Mingxi Zhu

Publications and source records attributed to Mingxi Zhu.

6 recordsLinked to original sources

arXiv2026

Mechanism for Collaborative Federated Learning: Pitfalls of Shapley Values

This paper investigates the impact of mechanism design on collaborative learning systems enabled by federated learning (FL). We propose a multi-action collaborative federated learning (MCFL) framework, capturing the interplay between agent strategies, platform mechanisms, and FL algorithms--a "three-body problem" in collaborative learning. This work demonstrates how the convergence rate and computational efficiency of FL are endogenously determined by the agent participation equilibrium that is induced by the mechanism. By doing so, we establish a direct link between incentive design in collaborative learning systems and the performance of the underlying optimization algorithms, a connection that has been largely overlooked in the existing literature. Specifically, we characterize the equilibrium of agent participation under two prominent mechanisms: the Shapley Value (SV) and Marginal Contribution (MC) mechanisms. Although SV is fair in surplus allocation and budget balanced, it has a vital pitfall: agents are incentivized to split their data across newly created fake identities. This is critical especially in the MCFL setting as it leads to slow convergence of FL optimization, which increases the number of required synchronization/communication rounds even when the per-round cost is fixed. In contrast, while MC is not budget-balanced, it is robust to such strategic manipulation and is able to induce an equilibrium that maximizes the MCFL system efficiency. Overall, our study lays a foundation for jointly designing incentives and algorithms in MCFL systems. We provide insights on pitfalls of SV: it induces a system equilibrium that leads to tremendous training cost and slower convergence, ultimately undermining the effectiveness of collaborative learning.

cs.GT↗

arXiv2025

Subspecialty-Specific Foundation Model for Intelligent Gastrointestinal Pathology

Gastrointestinal (GI) diseases represent a clinically significant burden, necessitating precise diagnostic approaches to optimize patient outcomes. Conventional histopathological diagnosis suffers from limited reproducibility and diagnostic variability. To overcome these limitations, we develop Digepath, a specialized foundation model for GI pathology. Our framework introduces a dual-phase iterative optimization strategy combining pretraining with fine-screening, specifically designed to address the detection of sparsely distributed lesion areas in whole-slide images. Digepath is pretrained on over 353 million multi-scale images from 210,043 H&E-stained slides of GI diseases. It attains state-of-the-art performance on 33 out of 34 tasks related to GI pathology, including pathological diagnosis, protein expression status prediction, gene mutation prediction, and prognosis evaluation. We further translate the intelligent screening module for early GI cancer and achieve near-perfect 99.70% sensitivity across nine independent medical institutions. This work not only advances AI-driven precision pathology for GI diseases but also bridge critical gaps in histopathological practice.

eess.IV↗

arXiv2025

How a Small Amount of Data Sharing Benefits Distributed Optimization and Learning : The Upside of Data Heterogeneity

Distributed optimization algorithms are widely used in machine learning. This paper investigates how a small amount of data sharing can improve their performance. Focusing on general linear models, we analyze the effects of data sharing on both primal and primal-dual optimization methods. Our contributions are threefold. First, from a theoretical perspective, we show that minimal data sharing improves algorithmic performance by shifting data from less favorable to more favorable structures. Contrary to the common belief that data heterogeneity is always harmful, we prove that while heterogeneity generally slows convergence in primal methods such as FedAvg and distributed PCG, it can accelerate convergence in primal-dual consensus algorithms like distributed ADMM, Fed-ADMM, and EXTRA by enriching dual dynamics. This reveals a form of duality in how heterogeneity affects different algorithm families. Second, building on this insight, we design a meta-algorithm for minimal data sharing, adaptable to both primal and primal-dual methods. We show that with as little as 1 percent shared data, convergence can be significantly accelerated across machine learning tasks. Finally, we argue from a broader perspective that even limited collaboration can yield large synergies, an idea that transcends the optimization context. Our findings provide both theoretical and practical guidance for improving distributed learning through minimal cooperation and motivate further exploration of cross-agent collaboration in solving complex global learning problems.

math.OC↗

arXiv2020

Sales Policies for a Virtual Assistant

We study the implications of selling through a voice-based virtual assistant (VA). The seller has a set of products available and the VA decides which product to offer and at what price, seeking to maximize its revenue, consumer- or total-surplus. The consumer is impatient and rational, seeking to maximize her expected utility given the information available to her. The VA selects products based on the consumer's request and other information available to it and then presents them sequentially. Once a product is presented and priced, the consumer evaluates it and decides whether to make a purchase. The consumer's valuation of each product comprises a pre-evaluation value, which is common knowledge, and a post-evaluation component which is private to the consumer. We solve for the equilibria and develop efficient algorithms for implementing the solution. We examine the effects of information asymmetry on the outcomes and study how incentive misalignment depends on the distribution of private valuations. We find that monotone rankings are optimal in the cases of a highly patient or impatient consumer and provide a good approximation for other levels of patience. The relationship between products' expected valuations and prices depends on the consumer's patience level and is monotone increasing (decreasing) when the consumer is highly impatient (patient). Also, the seller's share of total surplus decreases in the amount of private information. We compare the VA to a traditional web-based interface, where multiple products are presented simultaneously on each page. We find that within a page, the higher-value products are priced lower than the lower-value products when the private valuations are exponentially distributed. Finally, the web-based interface generally achieves higher profits for the seller than a VA due to the greater commitment power inherent in its presentation.

econ.TH↗

arXiv2020

Managing Randomization in the Multi-Block Alternating Direction Method of Multipliers for Quadratic Optimization

The Alternating Direction Method of Multipliers (ADMM) has gained a lot of attention for solving large-scale and objective-separable constrained optimization. However, the two-block variable structure of the ADMM still limits the practical computational efficiency of the method, because one big matrix factorization is needed at least once even for linear and convex quadratic programming. This drawback may be overcome by enforcing a multi-block structure of the decision variables in the original optimization problem. Unfortunately, the multi-block ADMM, with more than two blocks, is not guaranteed to be convergent. On the other hand, two positive developments have been made: first, if in each cyclic loop one randomly permutes the updating order of the multiple blocks, then the method converges in expectation for solving any system of linear equations with any number of blocks. Secondly, such a randomly permuted ADMM also works for equality-constrained convex quadratic programming even when the objective function is not separable. The goal of this paper is twofold. First, we add more randomness into the ADMM by developing a randomly assembled cyclic ADMM (RAC-ADMM) where the decision variables in each block are randomly assembled. We discuss the theoretical properties of RAC-ADMM and show when random assembling helps and when it hurts, and develop a criterion to guarantee that it converges almost surely. Secondly, using the theoretical guidance on RAC-ADMM, we conduct multiple numerical tests on solving both randomly generated and large-scale benchmark quadratic optimization problems, which include continuous, and binary graph-partition and quadratic assignment, and selected machine learning problems. Our numerical tests show that the RAC-ADMM, with a variable-grouping strategy, could significantly improve the computation efficiency on solving most quadratic optimization problems.

math.OC↗

arXiv2020

On a Randomized Multi-Block ADMM for Solving Selected Machine Learning Problems

The Alternating Direction Method of Multipliers (ADMM) has now days gained tremendous attentions for solving large-scale machine learning and signal processing problems due to the relative simplicity. However, the two-block structure of the classical ADMM still limits the size of the real problems being solved. When one forces a more-than-two-block structure by variable-splitting, the convergence speed slows down greatly as observed in practice. Recently, a randomly assembled cyclic multi-block ADMM (RAC-MBADMM) was developed by the authors for solving general convex and nonconvex quadratic optimization problems where the number of blocks can go greater than two so that each sub-problem has a smaller size and can be solved much more efficiently. In this paper, we apply this method to solving few selected machine learning problems related to convex quadratic optimization, such as Linear Regression, LASSO, Elastic-Net, and SVM. We prove that the algorithm would converge in expectation linearly under the standard statistical data assumptions. We use our general-purpose solver to conduct multiple numerical tests, solving both synthetic and large-scale bench-mark problems. Our results show that RAC-MBADMM could significantly outperform, in both solution time and quality, other optimization algorithms/codes for solving these machine learning problems, and match up the performance of the best tailored methods such as Glmnet or LIBSVM. In certain problem regions RAC-MBADMM even achieves a superior performance than that of the tailored methods.

math.OC↗