NSF-BSF: SHF: Small: Certifiable Verification of Large Neural Networks
NSF-BSF:SHF:小型:大型神经网络的可认证验证
基本信息
- 批准号:1814369
- 负责人:
- 金额:$ 48.09万
- 依托单位:
- 依托单位国家:美国
- 项目类别:Standard Grant
- 财政年份:2018
- 资助国家:美国
- 起止时间:2018-10-01 至 2022-09-30
- 项目状态:已结题
- 来源:
- 关键词:
项目摘要
Software systems play important roles in almost every area of modern life. In order to reduce the difficulty of developing new software, research in the field of artificial intelligence (AI) has been promoting a new model of programming: instead of having a human engineer design and code algorithms, a set of training examples are used together with machine-learning algorithms to automatically extrapolate software implementations. In classical programing, because such code is written by humans, we can persuade others that it is correct. In machine-learned systems, however, the program amounts to a highly complex mathematical formula for transforming inputs into outputs. The key difficulty, however, is that it is not possible currently to reason about correctness in such systems.This project addresses this issue by developing an algorithm, called Reluplex, capable of proving properties of deep neural networks (DNNs) or providing counter-examples if the properties fail to hold. The project has three main objectives. First, the investigators develop algorithmic techniques to greatly reduce the number of states that need to be explored by a verification tool. Second, they develop a strategy for producing checkable verification proofs. Checkable correctness proofs make it unnecessary to rely on correctness of the verification tool; one can instead rely only on the correctness of a small trusted proof-checker. Finally, the investigators implement this approach in an open-source tool and evaluate it on real-world industrial DNNs. Given that AI components are becoming ubiquitous in safety-critical systems, such as autonomous vehicles, this research will increase trust in these systems.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.
软件系统在现代生活的几乎每个领域都扮演着重要的角色。 为了降低开发新软件的难度,人工智能(AI)领域的研究一直在推动一种新的编程模式:不是让人类工程师设计和编码算法,而是将一组训练示例与机器学习算法一起使用,以自动推断软件实现。在经典编程中,因为这样的代码是由人类编写的,我们可以说服别人它是正确的。然而,在机器学习系统中,程序相当于将输入转换为输出的高度复杂的数学公式。然而,关键的困难是,目前无法在此类系统中推理正确性。该项目通过开发一种名为Reluplex的算法来解决这个问题,该算法能够证明深度神经网络(DNN)的属性,或者在属性不成立时提供反例。 该项目有三个主要目标。首先,研究人员开发算法技术,以大大减少需要由验证工具探索的状态的数量。 第二,他们开发了一种策略,用于生成可检查的验证证明。 可检查的正确性证明使得不需要依赖于验证工具的正确性;人们可以只依赖于一个小的可信的证明检查器的正确性。 最后,研究人员在开源工具中实现了这种方法,并在现实世界的工业DNN上对其进行了评估。鉴于人工智能组件在自动驾驶汽车等安全关键系统中变得无处不在,这项研究将增加对这些系统的信任。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(13)
专著数量(0)
科研奖励数量(0)
会议论文数量(0)
专利数量(0)
DeepCert: Verification of Contextually Relevant Robustness for Neural Network Image Classifiers
- DOI:10.1007/978-3-030-83903-1_5
- 发表时间:2021-03
- 期刊:
- 影响因子:0
- 作者:Colin Paterson;Haoze Wu;John M. Grese;R. Calinescu;C. Păsăreanu;Clark W. Barrett
- 通讯作者:Colin Paterson;Haoze Wu;John M. Grese;R. Calinescu;C. Păsăreanu;Clark W. Barrett
An SMT-Based Approach for Verifying Binarized Neural Networks
一种基于SMT的方法,用于验证二进制神经网络
- DOI:10.1007/978-3-030-72013-1_11
- 发表时间:2021-02-26
- 期刊:
- 影响因子:0
- 作者:Amir G;Wu H;Barrett C;Katz G
- 通讯作者:Katz G
Global optimization of objective functions represented by ReLU networks
- DOI:10.1007/s10994-021-06050-2
- 发表时间:2020-10
- 期刊:
- 影响因子:7.5
- 作者:Christopher A. Strong;Haoze Wu;Aleksandar Zelji'c;Kyle D. Julian;Guy Katz;Clark W. Barrett;Mykel J. Kochenderfer
- 通讯作者:Christopher A. Strong;Haoze Wu;Aleksandar Zelji'c;Kyle D. Julian;Guy Katz;Clark W. Barrett;Mykel J. Kochenderfer
Towards Verification of Neural Networks for Small Unmanned Aircraft Collision Avoidance
- DOI:10.1109/dasc50938.2020.9256616
- 发表时间:2020-10
- 期刊:
- 影响因子:0
- 作者:A. Irfan;Kyle D. Julian;Haoze Wu;Clark W. Barrett;Mykel J. Kochenderfer;Baoluo Meng;J. Lopez
- 通讯作者:A. Irfan;Kyle D. Julian;Haoze Wu;Clark W. Barrett;Mykel J. Kochenderfer;Baoluo Meng;J. Lopez
SAT Solving in the Serverless Cloud
无服务器云中的 SAT 解决方案
- DOI:10.34727/2021/isbn.978-3-85448-046-4_33
- 发表时间:2021
- 期刊:
- 影响因子:0
- 作者:Ozdemir, Alex;Wu, Haoze;Barrett, Clark
- 通讯作者:Barrett, Clark
{{
item.title }}
{{ item.translation_title }}
- DOI:
{{ item.doi }} - 发表时间:
{{ item.publish_year }} - 期刊:
- 影响因子:{{ item.factor }}
- 作者:
{{ item.authors }} - 通讯作者:
{{ item.author }}
数据更新时间:{{ journalArticles.updateTime }}
{{ item.title }}
- 作者:
{{ item.author }}
数据更新时间:{{ monograph.updateTime }}
{{ item.title }}
- 作者:
{{ item.author }}
数据更新时间:{{ sciAawards.updateTime }}
{{ item.title }}
- 作者:
{{ item.author }}
数据更新时间:{{ conferencePapers.updateTime }}
{{ item.title }}
- 作者:
{{ item.author }}
数据更新时间:{{ patent.updateTime }}
Clark Barrett其他文献
The nonexistence of unicorns and many-sorted L\"owenheim-Skolem theorems
独角兽的不存在和多种 L"owenheim-Skolem 定理
- DOI:
- 发表时间:
2024 - 期刊:
- 影响因子:0
- 作者:
Benjamin Przybocki;G. Toledo;Yoni Zohar;Clark Barrett - 通讯作者:
Clark Barrett
Being careful about theory combination
- DOI:
10.1007/s10703-012-0159-z - 发表时间:
2012-06-09 - 期刊:
- 影响因子:0.800
- 作者:
Dejan Jovanović;Clark Barrett - 通讯作者:
Clark Barrett
Efficiently Synthesizing Lowest Cost Rewrite Rules for Instruction Selection
有效综合用于指令选择的最低成本重写规则
- DOI:
- 发表时间:
2024 - 期刊:
- 影响因子:0
- 作者:
Ross G. Daly;Caleb Donovick;Caleb Terrill;J. Melchert;Priyanka Raina;Clark Barrett;Pat Hanrahan - 通讯作者:
Pat Hanrahan
Selected Extended Papers of NFM 2017: Preface
- DOI:
10.1007/s10817-018-9488-y - 发表时间:
2018-10-20 - 期刊:
- 影响因子:0.800
- 作者:
Clark Barrett;Temesghen Kahsai - 通讯作者:
Temesghen Kahsai
Clark Barrett的其他文献
{{
item.title }}
{{ item.translation_title }}
- DOI:
{{ item.doi }} - 发表时间:
{{ item.publish_year }} - 期刊:
- 影响因子:{{ item.factor }}
- 作者:
{{ item.authors }} - 通讯作者:
{{ item.author }}
{{ truncateString('Clark Barrett', 18)}}的其他基金
POSE: Phase II: An Open-Source Ecosystem for the cvc5 SMT Solver
POSE:第二阶段:cvc5 SMT 求解器的开源生态系统
- 批准号:
2303489 - 财政年份:2023
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF-BSF: SHF: Small: Neural Network Verification: Abstraction, Compositional Verification and Standardization
NSF-BSF:SHF:小型:神经网络验证:抽象、组合验证和标准化
- 批准号:
2211505 - 财政年份:2022
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF-BSF: SHF: Small: Efficient, Automatic, and Trustworthy Smart Contract Verification
NSF-BSF:SHF:小型:高效、自动且值得信赖的智能合约验证
- 批准号:
2110397 - 财政年份:2021
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
Collaborative Research: SHF: Small: Integrating Synthesis and Optimization in Satisfiability Modulo Theories
合作研究:SHF:小型:在可满足性模理论中集成综合和优化
- 批准号:
2006407 - 财政年份:2020
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF Student Travel Grant for 2019 Formal Methods in Computer-Aided Design (FMCAD)
NSF 2019 年计算机辅助设计形式方法 (FMCAD) 学生旅费补助金
- 批准号:
1935921 - 财政年份:2019
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
TWC: Medium: Collaborative: Breaking the Satisfiability Modulo Theories (SMT) Bottleneck in Symbolic Security Analysis
TWC:媒介:协作:打破符号安全分析中的可满足性模理论 (SMT) 瓶颈
- 批准号:
1228768 - 财政年份:2012
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
TC: EAGER: Collaborative Research: Parallel Automated Reasoning
TC:EAGER:协作研究:并行自动推理
- 批准号:
1049495 - 财政年份:2010
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
SHF: Small:Collaborative Research: Flexible, Efficient, and Trustworthy Proof Checking for Satisfiability Modulo Theories
SHF:小型:协作研究:灵活、高效且值得信赖的可满足性模理论证明检查
- 批准号:
0914956 - 财政年份:2009
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
相似国自然基金
枯草芽孢杆菌BSF01降解高效氯氰菊酯的种内群体感应机制研究
- 批准号:31871988
- 批准年份:2018
- 资助金额:59.0 万元
- 项目类别:面上项目
基于掺硼直拉单晶硅片的Al-BSF和PERC太阳电池光衰及其抑制的基础研究
- 批准号:61774171
- 批准年份:2017
- 资助金额:63.0 万元
- 项目类别:面上项目
B细胞刺激因子-2(BSF-2)与自身免疫病的关系
- 批准号:38870708
- 批准年份:1988
- 资助金额:3.0 万元
- 项目类别:面上项目
相似海外基金
Collaborative Research: NSF-BSF: How cell adhesion molecules control neuronal circuit wiring: Binding affinities, binding availability and sub-cellular localization
合作研究:NSF-BSF:细胞粘附分子如何控制神经元电路布线:结合亲和力、结合可用性和亚细胞定位
- 批准号:
2321481 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Continuing Grant
Collaborative Research: NSF-BSF: How cell adhesion molecules control neuronal circuit wiring: Binding affinities, binding availability and sub-cellular localization
合作研究:NSF-BSF:细胞粘附分子如何控制神经元电路布线:结合亲和力、结合可用性和亚细胞定位
- 批准号:
2321480 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Continuing Grant
NSF-BSF: Many-Body Physics of Quantum Computation
NSF-BSF:量子计算的多体物理学
- 批准号:
2338819 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Continuing Grant
Collaborative Research: NSF-BSF: Under Pressure: The evolution of guard cell turgor and the rise of the angiosperms
合作研究:NSF-BSF:压力之下:保卫细胞膨压的进化和被子植物的兴起
- 批准号:
2333889 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
Collaborative Research: NSF-BSF: Under Pressure: The evolution of guard cell turgor and the rise of the angiosperms
合作研究:NSF-BSF:压力之下:保卫细胞膨压的进化和被子植物的兴起
- 批准号:
2333888 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Continuing Grant
NSF-BSF: Towards a Molecular Understanding of Dynamic Active Sites in Advanced Alkaline Water Oxidation Catalysts
NSF-BSF:高级碱性水氧化催化剂动态活性位点的分子理解
- 批准号:
2400195 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF-BSF: Collaborative Research: Solids and reactive transport processes in sewer systems of the future: modeling and experimental investigation
NSF-BSF:合作研究:未来下水道系统中的固体和反应性输送过程:建模和实验研究
- 批准号:
2134594 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF-BSF Combinatorial Set Theory and PCF
NSF-BSF 组合集合论和 PCF
- 批准号:
2400200 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant
NSF-BSF: CDS&E: Tensor Train methods for Quantum Impurity Solvers
NSF-BSF:CDS
- 批准号:
2401159 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Continuing Grant
NSF-BSF: Collaborative Research: AF: Small: Algorithmic Performance through History Independence
NSF-BSF:协作研究:AF:小型:通过历史独立性实现算法性能
- 批准号:
2420942 - 财政年份:2024
- 资助金额:
$ 48.09万 - 项目类别:
Standard Grant