Michael Huang
Hi! I'm Michael. I am building Covalence Lab, where we work on human-ai interaction. I dropped out of mit, but previously I was a physics student and did an assortment of things semi-related to physics, cs, and ai. I was a competitive programmer in high school. In college, I worked on parallel algorithms under Prof. Julian Shun, did quantum photonics research under Prof. Soljacic, then worked on llm quantization at Intel's gpu team.
michaelyhuang23 at gmail dot com
Optimizers for Low Bit LLM Quantization
Intel Patent · Filed 2026
Developed gradient-based optimizer to perform low-bit quantization of LLM. Improved upon AutoRound's optimizer design by correcting for unstable dynamics caused by noisy STEs, achieving lower distortion.
Quantum Tunneling Dynamics of Biased OPO Systems
CLEO'25 · 2025
Analyzed biased OPO quantum state's stochastic dynamics with statistical physics. Approximated photon tunneling rate with a modified Kramer's method. Compared against numerical solution from finite element method.
Parallel Clustering with Graph Based Nearest Neighbor Search
ACDA'25 · 2025
Used approximate graph-based nearest neighbor search to accelerate high dimensional density-peaks clustering, achieving over 734x speedup over previous state-of-the-art.
Faster Parallel Exact Density Peaks Clustering
ACDA'23 · 2023
Developed the priority search kd-tree data structure, and applied it to a parallel density-peaks clustering algorithm that outperforms existing low-dimensional density-peaks clustering algorithms by up to 13000x.
Efficient Algorithms for Parallel Bi-core Decomposition
APoCS'23 · 2023
Developed efficient algorithms for bi-core decomposition, applied in fraudster detection and bioinformatics for analyzing bipartite graphs. Outperformed existing algorithms by up to 4.9x.