Sun Wins AAAI Best Paper Award

Jingwei Sun, a PhD student working in the laboratory of Yiran Chen, the John Cocke Distinguished Professor of Electrical and Computer Engineering at Duke University, won the “Best Paper Award” from the Federated Learning on the Edge Symposium, which is part of the Association for the Advancement of Artificial Intelligence Spring Symposium Series.

The paper titled “FedBPT: Efficient Federated Black-box Prompt Tuning for Large Language Models on the Edge,” was produced in collaboration with NVIDIA, a world leader in computer processing research, development and production.

Sun’s research introduces a method for training large language models, which power today’s generative artificial intelligence technologies like ChatGPT, on specific user data to fine-tune and personalize them without sharing the user’s data with a centralized entity. Called “Federated Learning,” this process allows users to reap the benefits of algorithms trained on large amounts of data while maintaining their own data’s privacy.

Crucially, Sun’s work does all of this in a way such that the data processing power requirements are low enough to be completed on users’ own mobile devices and local networks. It also facilitates model training without the user needing access to the model parameters, which is beneficial, especially considering that models are often black boxes to users in real-world applications.

“Experiments highlight our framework’s ability to drastically cut communication and memory costs while maintaining competitive performance,” Sun said. “Ultimately, FedBPT presents a promising solution for efficient, privacy-preserving fine-tuning of LLM in the age of large language models.”