Scholar
Zihang Liu
Google Scholar ID: SM_QghEAAAAJ
International Computer Science Institute, University of California, Berkeley
Citations & Impact
Citations
10
H-index
2
i10-index
0
Publications
2
Co-authors
5
list available
Publications
Resume (English only)
Academic Achievements
- Publications:
- - “LIFT the Veil for the Truth: Principal Weights Emerge after Rank Reduction for Reasoning-Focused Supervised Fine-Tuning”, ICML 2025
- - “Model Balancing Helps Low-data Training and Fine-tuning”, EMNLP 2024 (Oral Presentation)
- Conference Presentations:
- - November 2024, presentation on foundation model diagnosis at EMNLP 2024.
Research Experience
- Researcher @ UC Berkeley and International Computer Science Institute; started a research engineer position at ICSI, UC Berkeley, in June 2025, focusing on numerical algorithm discovery with deep learning.
Education
- Master Student @ UC Berkeley EECS, advised by Prof. Michael Mahoney; also works closely with Prof. Yaoqing Yang from Dartmouth College.
Background
- Research Interests: Understanding and improving the transparency and efficiency of learning models; particularly interested in low-rank structures, sparsity, and the geometry of weight matrices in deep learning models. Inspired by high-dimensional statistics, random matrix theory, and randomized linear algebra, also uses these techniques to discover new (numerical) algorithms.
