Computer Science Department, PhD Proposal Defense Yang Wu "Advancing the Proactive Capacity of LLMs in Cost-Sensitive Domains"

 Yang Wu  

PhD Candidate

WPI – Computer Science Department 

Date: May 5th, 2025

Time: 3pm-4pm 

Location: https://wpi.zoom.us/j/6065536440

Advisor: Prof. Xiaozhong Liu, PhD Advisor, WPI-Computer Science

Committee Members:

Prof. Kyumin Lee, WPI-Computer Science

Prof. Frank Zou, WPI- Mathematical Sciences

Prof. Haixu Tang, Indiana University Bloomington-Data Science

Abstract :

Large Language Models (LLMs) have demonstrated strong performance across a wide range of language tasks. However, their reactive nature—responding only when prompted by users—limits their ability to operate in scenarios that demand initiative and foresight. This dissertation focuses on enhancing the proactive capacity of LLMs through three key research questions that enable LLMs to act beyond prompt-driven interaction, particularly in cost-sensitive domains, e.g., legal and biomedical.

RQ1: Proactive Diagnostics for Knowledge Completion, which develops mechanisms enabling LLMs to autonomously detect incomplete information in user queries and proactively generate diagnostic questions, improving accuracy and robustness in knowledge-intensive scenarios such as legal consultation.

RQ2: Proactive Knowledge Acquisition under a Limited Budget, which investigates cost-effective strategies for LLMs to proactively select informative samples and appropriate domain experts for annotation, optimizing the efficiency of knowledge enrichment in scenarios constrained by annotation budgets and scarce expertise.

RQ3: Proactive Intervention in Streaming and Multimodal Settings, which explores how LLMs can proactively determine the appropriate timing and modality of intervention within streaming dialogues and real-world multimodal interactions, such as collaborative scientific discussions and interactive photo/video guidance.