Maximizing Knowledge Utility between Modelsand Data through Mutual Selection Strategies

The Hong Kong University of Science and Technology (Guangzhou)

数据科学与分析学域

PhD Thesis Proposal Examination

By Mr. LIU Hanmo

摘要

This thesis investigates how to maximize the utility of knowledge transfer between models and data through mutual selection strategies. We explore two complementary directions: (1) selecting suitable models for a given dataset by leveraging structured prior knowledge, and (2) selecting informative data subsets for training or replay under challenging conditions such as heterogeneity, unsupervised learning, and data evolution. We present methods including the Knowledge Benchmark Graph for data-to-model selection, and EDSR and ILEAHE for model-to-data selection. We maximally utilized the knowledge such that the efficiency of both building and updating models are highly improved while minimally sacrificing effectiveness. Together, these contributions advance a unified view of adaptive selection-based learning systems.

TPE Committee

Chair of Committee: Prof. CHU Xiaowen

Prime Supervisor: Prof. YANG Can

Co-Supervisor: Prof. CHEN Lei

Examiner: Prof. ZHANG Yongqi