EAGER: A Python Program Analysis Infrastructure to Facilitate Better Data Processing

EAGER：Python 程序分析基础设施，促进更好的数据处理

• 批准号: 1748764
• 负责人: Xiangyu Zhang
• 金额: $ 14.7万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-09-15 至 2019-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1748764&HistoricalAwards=false
• 关键词: EAGER; Python; Program; Analysis; Infrastructure

Python is the third most popular programming language, after C and Java,  and the most widely used language in Machine Learning and Data Science. Applications in Python are prone to human errors as much as those in other languages, or maybe more so due to the dynamic nature of Python. Therefore, tools to analyze, test, verify, and optimize Python applications are in a pressing need. Such tools are lagging or non-existent for Python. The root cause is the lack of infrastructure to support building practical and effective tools, which entails addressing the dynamic features of Python, such as dynamic typing, dynamic code loading/execution, and pervasive invocations to external library functions implemented in other languages. This project aims to explore the feasibility of building a Python program analysis infrastructure by developing two sample tools that rely upon a common set of infrastructural capabilities including the instrumentation, static analysis and symbolic analysis capabilities. The two sample tools are a data provenance tracking tool for machine learning applications and a bug finding tool to detect data format inconsistencies, which are the most dominant type of bugs in data processing. The provenance tool will demonstrate the importance of static analysis and program instrumentation, and the bug finding tool will demonstrate the importance of symbolic analysis. Both tools will illustrate the great benefits that can be brought to data scientists by advanced tools. In addition, they will illustrate that the aforementioned capabilities cannot be simply ported from existing infrastructures for other languages such as C and Java. The infrastructure will meet the pressing need of comprehensive tool building support for Python. A lot of cutting-edge synergistic research will be enabled across the CISE research community to serve data application programmers, data scientists and even end users.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.

Python是继C和Java之后第三大最受欢迎的编程语言，也是机器学习和数据科学中使用最广泛的语言。Python中的应用程序与其他语言中的应用程序一样容易出现人为错误，或者可能由于Python的动态特性而更容易出现人为错误。因此，迫切需要分析、测试、验证和优化Python应用程序的工具。这些工具对于Python来说是滞后的或不存在的。根本原因是缺乏支持构建实用和有效工具的基础设施，这需要解决Python的动态特性，例如动态类型，动态代码加载/执行，这个项目的目的是通过开发两个依赖于一组通用的基础设施能力，包括仪器，静态分析和符号分析能力。这两个示例工具是用于机器学习应用程序的数据来源跟踪工具和用于检测数据格式不一致的错误发现工具，这是数据处理中最主要的错误类型。起源工具将演示静态分析和程序插装的重要性，而错误发现工具将演示符号分析的重要性。这两种工具都将说明高级工具可以为数据科学家带来的巨大好处。此外，它们将说明上述功能不能简单地从其他语言（如C和Java）的现有基础设施中移植。该基础设施将满足对Python全面工具构建支持的迫切需求。许多尖端的协同研究将在整个CISE研究社区启用，以服务于数据应用程序员，数据科学家，甚至最终用户。该奖项反映了NSF的法定使命，并已被认为是值得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的支持。

项目成果

NIC: Detecting Adversarial Samples with Neural Network Invariant Checking

• DOI: 10.14722/ndss.2019.23415
• 发表时间: 2019
• 期刊: Proceedings 2019 Network and Distributed System Security Symposium
• 作者: Shiqing Ma;Yingqi Liu;Guanhong Tao;Wen-Chuan Lee;X. Zhang

ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation

• DOI: 10.1145/3319535.3363216
• 发表时间: 2019-11
• 期刊: Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security
• 作者: Yingqi Liu;Wen-Chuan Lee;Guanhong Tao;Shiqing Ma;Yousra Aafer;X. Zhang

LAMP: data provenance for graph based machine learning algorithms through derivative computation

• DOI: 10.1145/3106237.3106291
• 发表时间: 2017-08
• 期刊: Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering
• 作者: Shiqing Ma;Yousra Aafer;Zhaogui Xu;Wen-Chuan Lee;Juan Zhai;Yingqi Liu;X. Zhang