Scholar
Deqing Fu
Google Scholar ID: fsbgfqEAAAAJ
USC
Citations & Impact
Citations
284
H-index
7
i10-index
7
Publications
16
Co-authors
7
list available
Publications
Resume (English only)
Academic Achievements
- Publications:
- - Published papers on Transformer and in-context learning at NeurIPS 2024.
- - Presented work on decision theory for LLM reasoning under uncertainty at ICLR 2025, among others.
- - Contributed to research on modality sensitivity in Multimodal LLMs at COLM 2024.
- - Developed Token-level reward models (TLDR) for reducing hallucinations, presented at ICLR 2025.
- - Preprints: 'When Do Transformers Learn Heuristics for Graph Connectivity?' (arXiv, 2025), 'Zebra-CoT: A Dataset for Interleaved Vision Language Reasoning' (arXiv, 2025), 'Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models' (arXiv, 2025).
Research Experience
- Deqing conducts his doctoral research at USC, focusing on the understanding and improvement of large language models, as well as the development of methods for multimodal learning and synthetic data generation.
Education
- Undergraduate: Mathematics (with honors), University of Chicago; Master's: Statistics, University of Chicago; Ph.D.: Computer Science, University of Southern California, advised by Prof. Vatsal Sharan (USC Theory Group) and Prof. Robin Jia (Allegro Lab within USC NLP Group), and closely working with Prof. Mahdi Soltanolkotabi and Prof. Shang-Hua Teng.
Background
- Research Interests: Deep learning theory, natural language processing, and the interpretability of AI systems. Background: Deqing Fu is a fourth-year Ph.D. candidate in Computer Science at the University of Southern California (USC). His main research focus is on understanding large language models from algorithmic and theoretical perspectives, as well as developing methods for multimodal learning and synthetic data generation.
Miscellany
- Information on personal interests and hobbies not provided.
