Insoo Chung
Google Scholar · LinkedIn · GitHub · Resume
I'm an ML engineer interested in making neural networks run efficiently on small devices :)
The notes here are my views and my views only.
Now: At Meta, I work on low-bit quantization and on-device inference for Meta glasses, connecting numerical formats and kernels with hardware constraints. Earlier at Meta, I worked on reinforcement learning and LLM fine-tuning for audience-aware ad creation.
Previously: At Qualcomm AI Research, I worked on low-bit LLM optimization and evaluation. At Samsung Research, I worked on sub-3-bit transformer quantization and built training and C++ inference software for on-device translation. I also interned on Apple's Siri NL Platform, building an on-device benchmarking tool.
My research has appeared at EMNLP (co-first author), ACL (second author), WMT, and ICASSP. I studied computer science at Texas A&M University (M.C.S.) and Yonsei University (B.S.).
This website is my latest attempt at keeping a public notebook. I write (try to human-write) about whatever I'm currently trying to understand.
Email me at insoooo.chung@gmail.com for typos or corrections.