CICI: UCSS: Enhancing the Usability of Vulnerability Assessment Results for Open-Source Software Technologies in Scientific Cyberinfrastructure: A Deep Learning Perspective

联邦资助的科学网络基础设施(CI)加速了开创性的科学发现，包括黑洞成像、基因组测序、疫苗发现等。然而，有助于促进这些发现的开源软件(OSS)技术往往包含数千个漏洞，如果利用这些漏洞，可能会威胁到不可替代的科学分析。由于科学CI往往缺乏管理这些漏洞的人员，他们越来越多地将其漏洞管理任务外包给第三方研究和教育安全提供商，如OmniSOC。然而，这些提供商的安全分析员经常面临管理独联体开放源码软件资产中存在的数万个漏洞的挑战。该项目扫描数以千计的科学CI开放源码软件资产的漏洞，并采用新型的人工智能分析来(1)管理科学CI中的开放源码软件资产漏洞，(2)将它们与其补救战略联系起来。漏洞扫描和分析结果被集成到一个新的漏洞管理系统中，该系统允许安全分析师在跨科学CI的漏洞数据和补救策略上进行搜索、排序、浏览和协作。该项目设计了一个新的人工智能启用的AZSecure可用和协作科学安全框架，该框架跨两个主要的NSF资助的科学网络基础设施(CI)扫描四大类开源软件(OSS)资产(虚拟机、容器、基础设施即代码和GitHub)中的漏洞：(1)用于生命科学的CyVerse和(2)Jetstream，NSF为NSF和NIH提供的第一个科学和工程云。漏洞扫描支持三套人工智能分析研究推力，以增强漏洞扫描结果对OmniSOC安全分析师的可用性。第一种推力通过结合漏洞严重性加权方案和新的组合注意机制的多视角学习，将开放源码软件资产和漏洞数据聚合到漏洞管理任务的嵌入中。第二个推力使用自我监督学习和转换器，通过堆叠多个单词嵌入并将漏洞严重性分数与新的对比损失函数对齐，将漏洞扫描与修复策略联系起来。最后一个推力开发了一个漏洞管理系统，该系统集成了扫描结果，并使分析师能够操作这些方法。项目执行包括来自UArizona(NSA/DHS CD-，R和Co-Designed)和IU(NSA/DHS CD-and-R-Designed)的NSF CyberCorps服务奖学金研究生的角色。研究结果通过学术和行业出版物传播，并被整合到亚利桑那州大学和伊利诺伊州大学网络安全项目的顶级MS中。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Federally funded scientific cyberinfrastructure (CI) has accelerated ground-breaking scientific discoveries, including black hole imaging, genome sequencing, vaccine discovery, and more. However, the open-source software (OSS) technologies that help facilitate these discoveries often contain thousands of vulnerabilities that, if exploited, could threaten irreplaceable scientific analysis. Since scientific CIs often lack the personnel to manage these vulnerabilities, they increasingly outsource their vulnerability management tasks to third-party Research & Education security providers such as OmniSOC. However, security analysts at these providers often face challenges managing the tens of thousands of vulnerabilities present in OSS assets at CIs. This project scans thousands of scientific CI OSS assets for vulnerabilities and employs novel Artificial Intelligence-enabled analytics to (1) manage OSS asset vulnerabilities in scientific CI and (2) link them to their remediation strategies. Vulnerability scan and analytics results are integrated into a novel Vulnerability Management System that allows security analysts search, sort, browse, and collaborate on vulnerability data and remediation strategies across scientific CIs.This project designs a novel Artificial Intelligence-enabled AZSecure Usable and Collaborative Security for Science Framework that scans for vulnerabilities in four major categories of open-source software (OSS) assets (virtual machines, containers, infrastructure-as-code, and GitHub) across two major NSF-funded scientific cyberinfrastructures (CIs): (1) CyVerse for life sciences and (2) Jetstream, NSF’s first Science and Engineering Cloud for NSF and NIH. The vulnerability scans support three sets of AI-enabled analytics research thrusts to enhance the usability of vulnerability scan results for OmniSOC’s security analysts. The first thrust aggregates OSS asset and vulnerability data into an embedding for vulnerability management tasks through multi-view learning incorporating a vulnerability severity weighting scheme and a novel combinatorial attention mechanism. The second thrust uses self-supervised learning and transformers to link vulnerability scans with remediation strategies by stacking multiple word embeddings and aligning vulnerability severity scores with a novel contrastive loss function. The final thrust develops a Vulnerability Management System that integrates scan results and enables analysts to operate the methods. Project execution includes roles for NSF CyberCorps Scholarship-for-Service graduate students from UArizona (NSA/DHS CD-, R, and CO-designated) and IU (NSA/DHS CD- and- R-designated). Findings are disseminated through academic and industry publications and integrated into the top-ranked MS in Cybersecurity programs at UArizona and IU.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.