Scholar
Gregory Kang Ruey Lau
Google Scholar ID: MdVNNE8AAAAJ
National University of Singapore
Citations & Impact
Citations
115
H-index
5
i10-index
3
Publications
14
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- Papers: 'Position Paper: Uncover Scaling Laws for Large Language Models via Inverse Problems' accepted to Findings of EMNLP 2025; 'Dipper: Diversity in Prompts for Producing Large Language Model Ensembles in Reasoning tasks' accepted to EMNLP 2025; 'README: Rapid Equation Discovery with Multimodal Encoders' accepted to ICML 2025 AI4Math Workshop; 'Uncertainty Quantification for MLLM' accepted to ICML 2025 R2-FM'25 workshop; 'WaterDrum: Watermarking for Data-centric Unlearning Metric' accepted to ICML 2025 MUGen’25 workshop; 'PIED: Physics-Informed Experimental Design For Inverse Problems' accepted to AI4X 2025 conference (oral presentation); 'DUET: Optimizing Training Data Mixtures via Feedback from Unseen Evaluation Tasks' accepted to ICLR 2025 DATA-FM workshop; 'Uncertainty Quantification for MLLMs' accepted to ICLR 2025 QUESTION workshop; 'PIED: Physics-Informed Experimental Design for Inverse Problems' accepted to ICLR 2025. Awards: NUS School of Computing Research Achievement Award, EMNLP 2024 D&I Award.
Research Experience
- Before starting his Ph.D., he was a policymaker in the Singapore government, leading efforts in various areas such as data strategy, labor market policy, industry development, and social policy. He also spent some time as an entrepreneur, working on tech start-ups focused on education and career development.
Education
- Ph.D.: School of Computing, National University of Singapore, advised by Bryan Kian Hsiang Low, supported by the AI Singapore-CNRS@Create Descartes Joint PhD Scholarship; B.Sc.: Physics and Economics, Massachusetts Institute of Technology, worked with Wolfgang Ketterle, Eric Hudson, and Dave Donaldson; M.Fin.: MIT Sloan School of Management; MBA: Quantic School of Business and Technology.
Background
- Research Interests: Developing principled and practical methods to address data-centric challenges to improve modern AI systems. Themes include: (1) Selecting the right data to learn from in data-scarce regimes; (2) Tracing, auditing, and removing data influence in data-rich regimes; (3) Efficiently adapting pre-trained models at inference time without further training, and estimating their reliability.
Miscellany
- Interests: Cross-pollination of ideas across domains, such as AI, natural sciences, and social sciences.
Co-authors