ERI: Towards Intelligent, Cost-Efficient, and Adaptive Techniques to Enable Biomedical Hardware-Assisted Cybersecurity

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。几十年来，网络安全一直是全球关注的焦点，因为它是对社会的严重威胁，尤其是对国家的信息技术基础设施。医疗电子设备涵盖了一系列设备实现，从大型诊断成像设备到患者随身携带的小型移动设备。医疗保健系统中使用的现代计算平台的数字化特性及其与计算机网络的连接性增加导致了网络安全漏洞的增长，使此类系统成为复杂网络攻击的独特目标。拟议的项目旨在为新兴的医疗网络安全挑战开发有效和自适应的解决方案，以确保生物医学计算平台免受潜在的恶意网络攻击。此外，它还将产生综合工程研究项目和教育材料，以指导和培训人工智能（AI）和机器学习（ML）领域的本科生和研究生，用于安全分析和生物医学硬件辅助网络安全，特别是来自代表性不足的少数民族（URM）。拟议的工程研究和教育活动计划为来自不同工程背景（如计算机工程，计算机科学和生物医学工程专业）的学生提供必要的知识和技能，以便在要求苛刻的就业市场中具有竞争力，其中需要高度专业化的硬件设计师，人工智能/机器学习工程师，数据科学家和网络安全专家开发新的有效方法来保护现代计算系统。传统上，在计算系统中处理的数据的完整性在软件级别得到保护，假设底层硬件是安全的，不会受到潜在的攻击。然而，生物医学设备具有一些独特的特性，例如特定的安全要求、实现成本和设计权衡特征，以及破坏其安全协议的有限资源和计算能力。此外，严格的法规使得医疗计算机难以进行基本的软件更新，采用现成的反病毒（AV）保护也不足以防止恶意软件等新出现的网络攻击。为了克服传统的基于软件的解决方案的性能开销和低效率，现代生物医学设备中的安全性应该委托给底层硬件，构建一个自下而上的解决方案来保护计算设备，而不是将其视为事后的想法。在本项目中，我们将利用有效的AI/ML技术开发准确、低成本和自适应的技术，并为新兴生物医学设备中的硬件辅助网络安全构建多层智能框架。我们将利用生物医学设备中微处理器硬件组件捕获的低级硬件特征模式，构建新的智能技术，用于识别和分类新兴的网络攻击（例如，恶意软件，侧信道攻击，隐形攻击，零日攻击等），具有高精度和低计算开销。拟议的研究项目有五个主要目标：1)对生物医学设备处理器中新兴的硬件驱动网络攻击进行全面的数据收集、基准测试和特征分析；2)利用硬件相关特征为智能硬件辅助网络安全对策开发各种标准和先进的机器学习技术；3)探索智能硬件辅助对策的硬件实施结果和设备上权衡分析；4)开发基于强化学习的决策者，用于自适应选择准确且经济高效的检测器，以促进网络攻击的在线检测；5)开发基于系统级本体的网络安全分析框架，用于IoMT设备中有效的自动化知识推理。该奖项反映了美国国家科学基金会的法定使命，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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).For decades, cybersecurity has been at the forefront of global attention as a serious threat to society, particularly the nation's information technology infrastructure. Medical electronic devices cover a spectrum of equipment implementations, ranging from large diagnostic imaging machines to small mobile devices that patients carry with them. The digitized nature of modern computing platforms used in healthcare systems and their increased connectivity to computer networks have led into the growth of cybersecurity vulnerabilities making such systems a unique target for sophisticated cyber-attacks. The proposed project aims at developing effective and adaptive solutions for emerging healthcare cybersecurity challenges for securing biomedical computing platforms against potential malicious cyber-attacks. In addition, it will result in integrated engineering research projects and educational materials to mentor and train undergraduate and graduate students in the field of Artificial Intelligence (AI) and Machine Learning (ML) for security analysis and biomedical hardware-assisted cybersecurity specially from Under-Represented Minorities (URM). The proposed engineering research and educational activities are planned to provide students from diverse engineering backgrounds such as computer engineering, computer science, and biomedical engineering majors with the necessary knowledge and skills to be competitive in the demanding job market where highly specialized hardware designers, AI/ML engineers, data scientists, and cybersecurity specialist are needed to develop new efficient methods for securing modern computing systems. Traditionally, integrity of data being processed in computing systems has been safeguarded at the software level with the assumption that the underlying hardware is secure against potential attacks. However, biomedical devices have some unique features such as specific security requirements, implementation cost and design trade-offs characteristics, and limited resources and computational power that undermine their security protocols. In addition, strict regulations make it difficult to conduct basic software updates on medical computers and adopting off-the-shelf Anti-Virus (AV) protection is also insufficient for preventing emerging cyber-attacks such as malware. To overcome the performance overhead and inefficiency of conventional software-based solutions, the security in modern biomedical devices should be delegated to the underlying hardware, building a bottom-up solution for securing computing devices rather than treating it as an afterthought. In this project, we will leverage effective AI/ML techniques to develop accurate, low-cost, and adaptive techniques and build a multi-tiered intelligent framework for hardware-assisted cybersecurity in emerging biomedical devices. We will utilize the patterns of low-level hardware features captured by microprocessors’ hardware components in biomedical devices to build novel intelligent techniques for recognizing and classifying emerging cyber-attacks (e.g., malware, side-channel attacks, stealthy attacks, zero-day attacks, etc.) with high accuracy and low computational overheads. The proposed research project targets five major objectives: 1) Comprehensive data collection, benchmarking, and feature analysis of emerging hardware-driven cyber-attacks in biomedical devices’ processors, 2) Developing various standard and advanced machine learning techniques for intelligent hardware-assisted cybersecurity countermeasures using the hardware-related features, 3) Exploring hardware implementation results and on-device trade-off analysis of the intelligent hardware-assisted countermeasures, 4) Developing a reinforcement learning-based decision-maker for adaptive selection of the accurate and cost-efficient detector to facilitate online detection of the cyber-attacks, and 5) Developing a system-level ontology-based cybersecurity analysis framework for an effective automated knowledge reasoning in IoMT devices.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.