5/14/2024 Bruce Adams
Illinois Grainger Engineering Computer Science professors Nan Jiang, Dakshita Khurnana, Yupeng Zhang and Han Zhao have been named 2024 Google Research Scholars. Support from Google will aid their research in Large Language Models (LLMs), Cryptography, Privacy, Machine Learning, and Online and Offline Learning.
Written by Bruce Adams
In recent years, an influx of brilliant young scholars has been entering the University of Illinois Grainger Engineering Department of Computer Science faculty. Four CS professors have been named 2024 Google Research Scholars in three categories.
Machine learning and data mining
Han Zhao was chosen for the proposal “Trustworthy Machine Learning via Post-Processing.” He said that “this gift will allow our group to expand our current research agenda in the frontier of trust-worthy machine learning with a specific focus on promoting the ultimate fairness, robustness, and interpretability of current machine learning models, especially large language models. Our group particularly focuses on a paradigm called post-processing, which has many potential use cases for large language models. Many existing machine learning paradigms cannot be applied in the era of large language models because a small group of researchers, an academic group, does not have the data resources to retrain every model or have the computational capacity to do the retraining from scratch. However, our new paradigm will allow researchers to certify fairness, robustness, and interpretability constraints without retraining. The Google gift allows us to continue our research agenda for the next few years, and I look forward to future collaboration with researchers from Google on this important topic as well.”
Zhao’s research group comprises eight PhD students, two master’s students, and three undergraduates. He declared they are “top-notch students from their respective areas of interest, in this case, artificial intelligence and machine learning. I'm very much enjoying working with them. It's fair to say that without their help, it's not possible to achieve what we have achieved so far. I appreciate their participation in pushing for my research agenda.” In addition to the Google honor, Zhao was named a 2023 Kavli Fellow from the National Academy of Sciences, an invited speaker at the 2024 AAAI New Faculty Highlights program, and listed by CITL as an Excellent Teacher at the University of Illinois Urbana-Champaign in 2021-2023.
Nan Jiang was named for his proposal “Unification and Novel Interaction Protocols between Online and Offline RL.”
As Jiang explained, “My proposal is about bridging online and offline reinforcement learning; these are machine learning paradigms for decision making where the algorithms learn from trial and error in the real system (online) vs learn from pre-collected datasets (offline). They have been investigated extensively in parallel, but we are finding their deep connections that lead to potential unification. I am grateful for Google's recognition and looking forward to enhancing collaboration with some of their research teams.”
In addition to the Google honor, Jiang was named a 2024 Sloan Research Fellow, given a 2022 NSF Career Award, and listed by CITL as an Excellent Teacher at the University of Illinois Urbana-Champaign in 2022.
Privacy
Yupeng Zhang was selected for the proposal “Proof of Training and its Applications in Machine Unlearning and Differential Privacy.” Zhang described the paper’s subject matter: “Machine learning has seen prominent developments in recent years and is widely used in many applications. However, machine learning models are usually trained on large volumes of data, and data privacy is a rising concern. Users have no control over how their data is used in the training process. There is no mechanism to enforce any privacy protections, especially when the machine learning models are kept private.” Zhang suggests the use of “the cryptographic tool zero-knowledge proof, scaled to match the size of large language models (LLMs) via co-designed cryptography and machine learning training algorithms. Machine Unlearning is the concept advanced to remove private data from LLMs as required by recent regulations.” Zhang noted that “currently, there is no way to enforce this machine unlearning technique without monitoring the training process and learning the machine learning model. An important application of the proof of training in this proposal is to ensure that the data has been deleted without disclosing the machine learning model itself.”
As for his award, Zhang said, “It is a great recognition of my research. The award shows the importance of data privacy in machine learning technology to Google. It encourages this project and more research from the field to address the issues with cryptographic techniques.”
Quantum Computing
Dakshita Khurana was named for her proposal, “Cryptography for the Quantum Age.”
The Google Research Scholar Program aims to support early-career professors who are pursuing research in fields relevant to Google. It provides unrestricted gifts to support research at institutions worldwide and is focused on funding world-class research conducted by early-career professors. Awards are disbursed as unrestricted gifts of up to $60,000 USD to the university. They are intended for use during the academic year in which the award is provided to support the professor’s research efforts.