RGC Young Collaborative Research Grant (PC: Dr. Bo Han, Department of Computer Science, Hong Kong Baptist University)
Project Award Information
Award Number: RGC YCRG C2005-24Y Title: Towards Trustworthy Foundation Models under Imperfect Scenarios Principal Investigator (PC): Dr. Bo Han, Department of Computer Science, Hong Kong Baptist University
Project Summary
We have entered a new age of artificial intelligence, since Foundation Models (FMs) like ChatGPT and Sora emerge as pivotal tools with great capabilities in a broad range of domains and tasks. However, the deployment of FMs has surfaced critical concerns, particularly in robustness, safety, fairness, and reliability. In social science, while FMs offer advanced analysis of extensive qualitative data sets, they also face the problem of ensuring robustness against data anomalies and fairness in representation. In medical sciences, FMs promise a revolution through their ability to process large-scale medical datasets, yet they must do so with utmost safety and reliability to prevent harmful outcomes. Therefore, this project introduces solutions to the issues of FMs by developing trustworthy FMs. Specifically, trustworthy FMs will address the four grand challenges, including robustness against noisy inputs, safety against adversarial prompts, fairness against biased training data, and reliability against insufficient knowledge. Moreover, by developing advanced and targeted solutions, this project aims to bolster the functionality and dependability of trustworthy FMs, particularly within the critical spheres of social and medical sciences, thereby facilitating their responsible and beneficial integration into these fields. In summary, this collaborative project is expected to address the four grand challenges and construct trustworthy FMs, which can be further deployed to broader scientific and industrial applications.
Research Publications
The following papers focus on robustness against noisy inputs:
The following papers focus on safety against adversarial prompts:
The following papers focus on fairness against biased training data:
The following papers focus on reliability against insufficient knowledge:
Collaborators
University: TBD Institute: TBD Industry: TBD
Acknowlewdgement
This material is based upon work supported by the RGC under Grant No. C2005-24Y. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the RGC.
|