CIF: Small: A Simple and Unifying Optimization Framework for Signal and Information Processing Problems with Min-Max Structures

CIF：Small：针对具有最小-最大结构的信号和信息处理问题的简单且统一的优化框架

• 批准号: 1910385
• 负责人: Mingyi Hong
• 金额: $ 41万
• 依托单位: University of Minnesota-Twin Cities
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1910385&HistoricalAwards=false
• 关键词: CIF; Small; Simple; Unifying; Optimization

Over the past two decades, advances in signal and information processing have been heavily influenced by new developments in optimization. Modern optimization methods, from linear programming to convex optimization, have become indispensable tools to approach problems that arise in a number of core application areas such as compressive sensing, bioinformatics, wireless communication, and data analytics. Meanwhile, the tremendous growth in optimization algorithms for signal processing applications calls for the development of unifying frameworks, which can help simplify our understanding of algorithmic behavior, predict performance, and streamline the design of application-specific algorithms. The framework developed in this project will benefit a wide range of applications well beyond signal processing, including machine learning, data mining and computer vision. The proposed efforts also offer rich and varied opportunities for engaging undergraduate students in cross-disciplinary research, as well as in K12 outreach activities. This project is focused on designing a framework that deals with a challenging class of block Minimization-Maximization problems, which involves jointly minimizing and maximizing a particular class of structured objective functions. This covers many signal processing applications, old and new, such as min rate utility maximization and robust data analytics. This project will take a bottom-up approach, where starting from a relatively simple formulation, problem-specific features and properties will be gradually built. Rigorous performance analysis will be conducted, and the resulting algorithms will be specialized and evaluated in a number of signal processing applications, including wireless resource allocation in the presence of jammers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

在过去的二十年里，信号和信息处理的进步受到了优化新发展的重大影响。现代优化方法，从线性规划到凸优化，已经成为解决压缩传感、生物信息学、无线通信和数据分析等核心应用领域中出现的问题的不可或缺的工具。同时，信号处理应用优化算法的巨大增长要求开发统一的框架，这可以帮助简化我们对算法行为的理解，预测性能，并简化特定于应用的算法的设计。该项目开发的框架将使信号处理以外的广泛应用受益，包括机器学习，数据挖掘和计算机视觉。拟议的努力还提供了丰富多样的机会，让本科生参与跨学科研究，以及在K12外展活动。该项目的重点是设计一个框架，处理具有挑战性的块最小化-最大化问题，其中涉及联合最小化和最大化一类特定的结构化目标函数。这涵盖了许多新老信号处理应用，例如最小速率效用最大化和强大的数据分析。该项目将采取自下而上的方法，从相对简单的公式开始，逐步建立针对具体问题的功能和属性。将进行严格的性能分析，并将产生的算法专门化，并在一些信号处理应用中进行评估，包括存在干扰机的无线资源分配。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Finding First-Order Nash Equilibria of Zero-Sum Games with the Regularized Nikaido-Isoda Function

• 发表时间: 2021
• 作者: Ioannis C. Tsaknakis;Mingyi Hong

Revisiting and Advancing Fast Adversarial Training Through The Lens of Bi-Level Optimization

• 发表时间: 2021-12
• 作者: Yihua Zhang;Guanhua Zhang;Prashant Khanduri;Min-Fong Hong;Shiyu Chang;Sijia Liu

Hybrid Block Successive Approximation for One-Sided Non-Convex Min-Max Problems: Algorithms and Applications

• DOI: 10.1109/tsp.2020.2986363
• 发表时间: 2019-02
• 期刊: IEEE Transactions on Signal Processing
• 影响因子: 5.4
• 作者: Songtao Lu;Ioannis C. Tsaknakis;Mingyi Hong;Yongxin Chen

Distributed Adversarial Training to Robustify Deep Neural Networks at Scale

• DOI: 10.48550/arxiv.2206.06257
• 发表时间: 2022-06
• 作者: Gaoyuan Zhang;Songtao Lu;Yihua Zhang;Xiangyi Chen;Pin-Yu Chen;Quanfu Fan;Lee Martie;L. Horesh

Min-Max Optimization without Gradients: Convergence and Applications to Black-Box Evasion and Poisoning Attacks

• 发表时间: 2020-07
• 作者: Sijia Liu;Songtao Lu;Xiangyi Chen;Yao Feng;Kaidi Xu;Abdullah Al-Dujaili;Mingyi Hong;Una-May O’Reilly