Chiao Lun Cheng
About Chiao Lun Cheng
Chiao Lun Cheng is a Founding ML Engineer at Scale AI, specializing in large language models and machine learning systems. He has a strong academic background with a PhD from MIT and extensive experience in various technology roles across multiple companies.
Work at ScaleAI
Chiao Lun Cheng has been serving as a Founding ML Engineer at Scale AI since 2018. In this role, he has focused on developing advanced machine learning solutions, including creating model training and deployment systems that leverage Kubernetes and NATS.IO service bus technology. His work enables autoscaling capabilities without session losses, enhancing the efficiency of machine learning operations.
Education and Expertise
Chiao Lun Cheng holds a PhD in Physical Chemistry and Computational Physics from the Massachusetts Institute of Technology, where he studied from 2003 to 2008. He also earned a B.S. in Chemistry, Biology from UC Irvine and a B.S. in Chemistry, Physics, and Biochemistry from the University of California, Berkeley. His academic background provides a strong foundation in scientific principles and computational techniques relevant to machine learning.
Background
Chiao Lun Cheng has a diverse professional background, having worked in various roles across multiple organizations. He was a Research Assistant at the Massachusetts Institute of Technology from 2003 to 2008, followed by an Associate position at McKinsey & Company from 2008 to 2010. He then served as Principal at Katong Capital from 2010 to 2014 and as Chief Technology Officer at Minus Inc. from 2014 to 2015.
Technical Contributions
Chiao Lun Cheng has made significant technical contributions throughout his career. He has finetuned large language models using 8-bit base models and 16-bit LoRA adapters. His work includes implementing opportunistic batching of inference using the CUDA Streams synchronization API and developing retrieve and rerank pipelines for LLM-based Q&A systems. He has also optimized data loading for training from S3 stores through profiling techniques.
Research and Development Initiatives
Chiao Lun Cheng has engaged in various research and development initiatives, including modeling railway arrival times using word2vec and mixture density networks. He has applied metric learning for fashion styles through non-negative matrix factorization and utilized Item Response Theory to model labeler accuracy and task difficulty. His work often integrates advanced machine learning techniques to solve complex problems.