Over the years, and especially recently, the accelerated development of new machine learning technologies has captured the attention of security- and privacy-focused researchers.
Of course, these advances and their vulnerabilities to AI applications leave users open to attack.
In response, Illinois Computer Science Professor Bo Li emphasized robustness, privacy, generalization, and the underlying interconnections of these items, paving the way for a career as a researcher into a credible field of machine learning. located at the intersection.
“As we are increasingly aware, machine learning is now ubiquitous in the technology world through areas ranging from autonomous driving, large language models, ChatGPT, and more,” Lee said. rice field. “This is an advantage that we also see in various applications, such as facial recognition technology.
“The troubling aspect is that we’ve also learned the threat that these advances are vulnerable to attack.”
Lee ‘happy’ to be on IEEE AI 10 watchlist
Earlier this month, Bo Li logged on to her computer and noticed several emails congratulating her from colleagues and students.
But she didn’t know exactly what it was for.
“After all, I recently learned that Twitter is the way we find so much information,” Li said with a laugh.
There she saw several notices stemming from the IEEE’s announcement of AI 10 to the 2022 watch list. It also included her name.
“It was great to hear such wonderful words from current and past students and collaborators, and I am happy to be on this meaningful list for IEEE,” Lee said.
The Illinois CS professor’s early career in academia earned him awards such as the IJCAI Computers and Thought Award, the Alfred P. Sloan Research Fellowship, the NSF CAREER Award, the IEEE’s Top 10 list for AI, and the MIT Technology Review TR-35. , has already earned considerable honors.awards, etc.
Li’s work includes research awards from technology companies such as Amazon, Meta, Google, Intel, MSR, eBay, and IBM, as well as best paper awards at several major machine learning and security conferences.
“Each recognition and award represents a great deal of support for my research, and each gave me confidence in the direction I’m working in,” Lee said. “I am very happy and grateful to all the nominators and the community. Any and all evaluations, including the IEEE AI 10 to Watch List, are a very interesting and important signal that my work is valuable to different people in different ways.” gives me the
IEEE Intelligent Systems Editor-in-Chief San Murugesang emphasized the importance of this year’s winners, who are rising stars in what he calls the Golden Age of AI, offering incredible opportunities.
Lee thanked his mentor here at Illinois CS, Professor David Forsyth, and influences from his time at UC Berkeley, such as Dawn Song and Stuart Russell.
Their steady guidance has set her up for a successful early academic career. And Lee is ready to give back to the next generation of influential AI scholars.
“My first advice is to read a lot of good literature and talk to seniors you respect, so develop your own deep and unique research tastes,” Lee said. . “Good researchers provide unique and deep insights. It’s rare, and it takes a lot of effort. But the work is worth it.”
Already having a successful start to his problem-focused career, Lee made $1 million to align his secure learning lab with DARPA’s Guaranteed AI Robustness Against Fraud (GARD) program.
She said the project is purely for research purposes. Divide the stakeholders into different teams. The red team presents vulnerabilities and attacks that the blue team tries to defend against.
The organizers believe this vulnerability is too complex to be resolved during the life of this project, but the value of the work goes beyond just fixing the vulnerability.
“For students coming from my lab, this is an opportunity to tackle ambitious projects without the pressure of leaderboards and competitive end results,” Lee said. “This is ultimately an assessment that helps us understand the algorithms involved. Because it is open source and consists of coherent meetings, everyone works together to uncover progress and make it the most I can understand it very well.”
The ultimate goal for both her and her students is to define this threat model in a more precise way.
“We cannot say that our systems or machine learning systems can be trusted against arbitrary attacks. Told. “And then you have to define what you trust. For example, given a task, given a dataset that provides a model, there is this different specific requirement.
“And by optimizing the end-to-end system, we will be able to guarantee the metrics we care about. We hope to be able to offer a more rigorous warranty in the future.”
This is a continuation of Lee’s years of research with his students towards the concept of trustworthy AI.
For example, previous breakthroughs looked at consistent give-and-take between components that create trustworthy AI.
Researchers felt that there had to be a certain trade-off between accuracy and robustness for systems that deal with machine learning vulnerabilities.
But Lee and her group said they proposed a framework called “learning reasoning” that integrates human reasoning into the equation to mitigate such tradeoffs.
“What we are aiming for is a scenario where AI developers understand that it is important to prioritize both robustness and accuracy or safety at the same time,” Li said. “Many times the process simply puts performance first. When that happens, organizers worry about protecting it later. I think it will help in the proper development of our technology.”
Further research by her students has resulted in advancements in related fields.
For example, Ph.D. student Linyi Li built an integrated toolbox that provides certified robustness to deep neural networks.
Additionally, Ph.D. student Chejian Xu and master’s student Jiawei Zhang created various safety-critical scenarios for self-driving cars. They plan to hold his CVPR workshop on it in June.
Finally, Zhang and Ph.D. student Mintong Kang together built a scalable learning reasoning framework.
Such developments also led to Lee’s involvement in the newly formed NSF AI Institute for Agent-based Cyber Threat Intelligence and OperationoN (ACTION).
Led by the University of California, Santa Barbara, the NSF ACTION Institute also aims to revolutionize the protection of mission-critical systems against advanced cyberthreats.
“Among the most impactful potential outcomes from the ACTION Institute are various fundamental algorithms for AI and cybersecurity and large-scale AI-enabled systems for cybersecurity tasks with formal security assurances. These include not only purely data-driven models, but also domain knowledge, weak human oversight, and directed logical reasoning. It also applies to unsolicited attacks,” Lee said.
Despite the rapid pace of development in AI and machine learning, it’s clear that Li and her students are dedicated to stabilizing and securing this process going forward.
