权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

III: Small: Combinatorial Algorithms for High-dimensional Learning

III：小：高维学习的组合算法

基本信息

批准号：
2008557
负责人：
Andreas Stathopoulos
金额：
$ 39.54万
依托单位：
College of William and Mary
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2020
资助国家：
美国
起止时间：
2020-10-01 至 2024-09-30
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=2008557&HistoricalAwards=false
关键词：
III Small Combinatorial Algorithms dimensional

项目摘要

This project designs a new suite of statistical learning algorithms for high-dimensional problems in which the number of observations is insufficient to support conventional methods. Solving these problems is vital for modern scientific discoveries in social and environmental issues that rely on both complex models and massive data. For example, many empirical asset pricing models can use only 10 years of daily market data, for a total of approximately 3,000 observations, to fit significantly more than 3,000 parameters. The algorithms created in the project integrate key tools from theoretical computer science with data to deliver more accurate predictions with significantly fewer samples. The suite of new algorithms will assist with data-centric problems in financial econometrics, social network analysis, recommender systems, and skillset inferences. From a technical standpoint, this project uses graph-based algorithms and average case analysis to exploit statistical patterns that cannot be addressed by existing tools. These patterns include, for instance, block diagonal structures of the learnable parameters in vector regression models, and covariance matrices of mildly correlated features that exhibit heavy-tailed spectra. The project consists of two thrusts. First is the design of learning algorithms by relating high-dimensional problems to graph-learning problems, and generalizing graph-learning techniques. Second is the construction of new algorithms based on average case analysis, motivated by (i) the avoidance of over-conservativeness by using distributional assumptions, and (ii) the use of ``promises'', a notion borrowed from theoretical computer science, that guides prediction power using input structures. Utilizing weak distributional assumptions and promises together in turn allows the design of effective algorithms for a multitude of problems.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.

该项目设计了一套新的统计学习算法，用于高维问题，其中观察的数量不足以支持传统的方法。解决这些问题对于依赖复杂模型和大量数据的社会和环境问题的现代科学发现至关重要。例如，许多经验资产定价模型可以仅使用10年的每日市场数据，总共约3，000次观察，以拟合超过3，000个参数。该项目中创建的算法将理论计算机科学的关键工具与数据相结合，以显著减少的样本提供更准确的预测。这套新算法将有助于解决金融计量经济学、社交网络分析、推荐系统和技能推理中以数据为中心的问题。从技术角度来看，该项目使用基于图形的算法和平均案例分析来利用现有工具无法解决的统计模式。例如，这些模式包括向量回归模型中可学习参数的块对角结构，以及表现出重尾谱的轻度相关特征的协方差矩阵。该项目包括两个重点。首先是通过将高维问题与图学习问题联系起来设计学习算法，并推广图学习技术。其次是基于平均案例分析的新算法的构建，其动机是（i）通过使用分布假设来避免过度保守，以及（ii）使用“承诺”，这是一个从理论计算机科学中借来的概念，它使用输入结构来指导预测能力。该奖项反映了NSF的法定使命，并通过使用基金会的知识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

期刊论文数量（11）

专著数量（0）

科研奖励数量（0）

会议论文数量（0）

专利数量（0）

BATS: A Spectral Biclustering Approach to Single Document Topic Modeling and Segmentation

DOI：
10.1145/3468268
发表时间：
2021-12-01
期刊：
ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY
影响因子：
5
作者：
Wu, Qiong;Hare, Adam;Li, Yanhua
通讯作者：
Li, Yanhua

Towards Socially Acceptable Food Type Recognition

DOI：
10.1109/msn57253.2022.00110
发表时间：
2022-12
期刊：
2022 18th International Conference on Mobility, Sensing and Networking (MSN)
影响因子：
0
作者：
Junjie Wang;Jiexiong Guan;Y.Alicia Hong;†. HongXue;Shuangquan Wang;Zhenming Liu;Bin Ren;Gang Zhou;William Mary
通讯作者：
Junjie Wang;Jiexiong Guan;Y.Alicia Hong;†. HongXue;Shuangquan Wang;Zhenming Liu;Bin Ren;Gang Zhou;William Mary

iQAN: Fast and Accurate Vector Search with Efficient Intra-Query Parallelism on Multi-Core Architectures

DOI：
10.1145/3572848.3577527
发表时间：
2023-02
期刊：
Proceedings of the 28th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming
影响因子：
0
作者：
Zhen Peng;Minjia Zhang;K. Li;R. Jin;Bin Ren
通讯作者：
Zhen Peng;Minjia Zhang;K. Li;R. Jin;Bin Ren

Parallel Software for Million-scale Exact Kernel Regression

DOI：
10.1145/3577193.3593737
发表时间：
2023-06
期刊：
Proceedings of the 37th International Conference on Supercomputing
影响因子：
0
作者：
Yu Chen;Lucca Skon;James R. McCombs;Zhenming Liu;A. Stathopoulos
通讯作者：
Yu Chen;Lucca Skon;James R. McCombs;Zhenming Liu;A. Stathopoulos

Equity2Vec: end-to-end deep learning framework for cross-sectional asset pricing

DOI：
10.1145/3490354.3494409
发表时间：
2019-09
期刊：
Proceedings of the Second ACM International Conference on AI in Finance
影响因子：
0
作者：
Qiong Wu;Christopher G. Brinton;Zhenghao Zhang;A. Pizzoferrato;Zhenming Liu;Mihai Cucuringu
通讯作者：
Qiong Wu;Christopher G. Brinton;Zhenghao Zhang;A. Pizzoferrato;Zhenming Liu;Mihai Cucuringu