CAREER: Advances in Graph Learning and Inference

职业：图学习和推理的进展

• 批准号: 2005804
• 负责人: Chinmay Hegde
• 金额: $ 36.47万
• 依托单位: New York University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-11-01 至 2024-01-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2005804&HistoricalAwards=false
• 关键词: CAREER; Advances; Graph; Learning; Inference

Graph-based data processing algorithms impact a variety of application domains ranging from transportation networks, artificial intelligence systems, cellphone networks, social networks, and the Web. Nevertheless, the emergent big-data era poses key conceptual challenges: several existing graph-based methods used in practice exhibit unreasonably high running time; several other methods operate in the absence of correctness guarantees. These challenges severely imperil the safety and reliability of higher-level decision-making systems of which they are a part. This research introduces an innovative new computational framework for graph learning and inference that addresses these challenges. Specific applications studied in this project include: better approaches for monitoring roadway congestion and identify traffic incidents in a timely manner; root-cause analysis of complex events in social networks; and design of better personalized learning systems, lowering educational costs and increasing quality nationwide. Activities include integrated programs to increase participation of women and under-represented minorities in the computational sciences. From a technical standpoint, the investigator pursues three research themes: (i) designing scalable non-convex algorithms for learning the edges (and weights) of an unknown graph given a sequence of independent static and/or time-varying local measurements; (ii) designing new approximation algorithms for utilizing the structure of a given graph to enable scalable post-hoc decision making in complex systems; (iii) developing provable algorithms for training special families of artificial neural networks, and filling gaps between rigorous theory and practice of neural network learning. Progress in each of the above themes will be extensively evaluated using real-world data from engineering applications including social network data, highway monitoring data, and fluid-flow simulation data. Collaborations with domain experts in each of these application areas will ensure that the new theory, tools, and software emerging from this project will lead to meaningful societal benefits.

基于图的数据处理算法影响着交通网络、人工智能系统、手机网络、社交网络和网络等各种应用领域。然而，新兴的大数据时代带来了关键的概念挑战：实践中使用的几种现有的基于图的方法表现出不合理的高运行时间；其他几种方法在缺乏正确性保证的情况下运行。这些挑战严重危及它们所属的高层决策系统的安全性和可靠性。这项研究引入了一种用于图学习和推理的创新计算框架，可以解决这些挑战。该项目研究的具体应用包括：更好的方法来监控道路拥堵并及时识别交通事件；社交网络中复杂事件的根本原因分析；设计更好的个性化学习系统，降低教育成本并提高全国范围内的质量。活动包括提高妇女和代表性不足的少数群体对计算科学的参与的综合计划。从技术角度来看，研究者追求三个研究主题：（i）设计可扩展的非凸算法，用于在给定一系列独立静态和/或时变局部测量的情况下学习未知图的边缘（和权重）； (ii) 设计新的近似算法，利用给定图的结构来实现复杂系统中可扩展的事后决策； (iii) 开发可证明的算法来训练特殊的人工神经网络系列，并填补神经网络学习的严格理论与实践之间的空白。上述每个主题的进展将使用来自工程应用的真实数据进行广泛评估，包括社交网络数据、高速公路监测数据和流体流动模拟数据。与每个应用领域的领域专家的合作将确保该项目中出现的新理论、工具和软件将带来有意义的社会效益。

项目成果

Implicit Regularization for Group Sparsity

群稀疏性的隐式正则化

• 发表时间: 2023-07
• 期刊: International Conference on Learning Representations
• 作者: Li; J. and
• 通讯作者: J. and

Fast and Provable Algorithms for Learning Two-Layer Polynomial Neural Networks

用于学习两层多项式神经网络的快速且可证明的算法

• DOI: 10.1109/tsp.2019.2916743
• 发表时间: 2019-07
• 期刊: IEEE Transactions on Signal Processing
• 影响因子: 5.4
• 作者: Soltani, Mohammadreza;Hegde, Chinmay
• 通讯作者: Hegde, Chinmay

Reducing the Search Space for Hyperparameter Optimization Using Group Sparsity

使用组稀疏性减少超参数优化的搜索空间

• DOI: 10.1109/icassp.2019.8682434
• 发表时间: 2019-04-24
• 期刊: ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
• 作者: Minsu Cho;C. Hegde
• 通讯作者: C. Hegde

Benefits of Jointly Training Autoencoders: An Improved Neural Tangent Kernel Analysis

联合训练自动编码器的好处：改进的神经切线核分析

• DOI: 10.1109/tit.2021.3065212
• 发表时间: 2021-07
• 期刊: IEEE Transactions on Information Theory
• 影响因子: 2.5
• 作者: Nguyen, Thanh V.;Wong, Raymond K.;Hegde, Chinmay
• 通讯作者: Hegde, Chinmay

InvNet: Encoding Geometric and Statistical Invariances in Deep Generative Models

InvNet：在深度生成模型中编码几何和统计不变性

• DOI: 10.1609/aaai.v34i04.5863
• 发表时间: 2020-04-03
• 期刊: Industrial & Engineering Chemistry Research
• 作者: Ameya Joshi;Minsu Cho;Viraj Shah;B. Pokuri;S. Sarkar;B. Ganapathysubramanian;C. Hegde
• 通讯作者: C. Hegde