Seoul, South Korea
Kian (Dahyun) Kim is currently an ML Research Manager at TwelveLabs, leading the Pegasus team. Prior to the current managerial role, he was one of the key individual contributors of Pegasus, an in-house video multi-modal foundation model from Twelve Labs and the first ever video multi-modal model to be hosted in AWS Bedrock. Before joining Twelve Labs, he worked as an AI Research Engineer at Upstage, where he played an instrumental role in reaching world #1 spot in the Open LLM Leaderboard with the Solar 10.7B model. Following the above, he also worked on building next-generation LLMs at Upstage such as the Solar-pro-preview. He completed both B.S and M.S degrees with summa cum laude from Gwangju Institute of Science and Technology (GIST) under the supervision of Prof. Jonghyun Choi (who is now an associate professor at SNU). He is quick to adapt and excel in multiple subfields of AI, including but not limited to, multi-modal foundation models, LLMs, OCR, binary networks, and continual learning.
Learning Architectures for Binary Networks, ECCV2020 바이너리 뉴럴 네트워크를 위한 뉴럴 아키텍쳐 서치 방법 및 장치, KR Patent Method and device for neural architecture search optimized for binary neural network, US Patent BNAS-v2: A Summary with Empirical Improvements, CVPR2021 Workshop Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime Inference, ICLR2022 Unsupervised Representation Learning for Binary Networks by Joint Classifier Learning, CVPR2022