Kevin Miao
Home

My name is Kevin Miao! I am currently working at Apple's Vision Products Group as a Machine Learning Research Scientist. My work involves developing/researching novel techniques for Vision-based foundation models and prototyping spatial applications & tooling based on those models.


my work

My research lies at the intersection of explainable artificial intelligence, multimodal representation learning and 3D diffusion. Generally, I am interested in investigating how we can make such machine learning models more emergent, natural to use and powerful. Rather than focusing on changing up architecture and increasing on benchmarking efficiency, I am determined to find out how we can learn better representations using different data modalities (text, 2D, 3D, video, senors, etc.) and training techniques. For instance, how do we efficiently improve 3D synthesis by combining text, 2D and 3D data? Or how do we pretrain/finetune multimodals to create a semantic search model? How do we ensure that these models are well-aligned and robust against perturbations or distribution shifts? In addition to devising novel machine learning techniques and applications, I have also closely collaborated with Visual Analytics and HCI (Human-Computer Interaction) experts on how to create useful interfaces and interaction techniques for ML developers on interpreting and explaining their models.

more background (toggle) πŸ‘ˆ

Before Apple, I finished my Master's degree in Electrical Engineering and Computer Science ('22) at UC Berkeley. 🐻 Under the supervision of Joey Gonzalez, Trevor Darrell and Kurt Keutzer. In my thesis work, I devised a framework for knowledge-embedded regularization in self-supervised vision transformers (Arxiv link can be found below). Prior to that I finished my undergraduate degree at UC Berkeley in Computer Science with honors. However, my original academic interests have actually always been in Biology and Medicine until my third year of college when I first learned to code. I investigated the evolution of the olfactory neurons in S. Flava (a fruit fly species) under Noah Whiteman. After switching my major to Data Science (and then to Computer Science), I combined my biomedical and computative knowledge to investigate and develop precision medicine frameworks at UCSF with Julian Hong. In the summer after obtaining my bachelor's degree in 2021, I worked as a Machine Learning Intern at Felyx, an Amsterdam-based mobility startup, and created an end-to-end computer vision pipeline to detect wrongly parked vehicles.

publications

talks

  • May 2024 at Stanford XR. Exploring the Synergy: AI in Extended Reality
  • February 2024 at UC Berkeley Data Science Society. The Era After Big Data and Large Language Models: Building Generalist Agents
  • October 2023 at UC Berkeley. A Story on Data-Centric Machine Learning

teaching

Continuing my passion for teaching students state-of-the-art skills and insights in data science, I am excited to be part of UC Berkeley's brand new College of Computing, Data Science and Society as a lecturer. Here below, I detail the courses I teach and have taught. All course notes and other resources are publicly available where possible. My belief is that education should be equitable; resources and materials should not be hidden away from the public. I am fortunate enough to have taught and be teaching courses that share this philosophy.

Data 100: Principles and Techniques of Data Science
Data 8: The Foundations of Data Science
Data 198-003: Data Discovery Scholars Research Seminar

mentoring

  • Michelle Chang (2023; Apple; Now at Harvard MS Data Science)
  • Jonathan Ferrari (2024; UC Berkeley)
  • Cindy Yang (2020-2021; UC Berkeley; Now at UC Berkeley MS IEOR)
  • Lydia Sidhom (2022; UC Berkeley)
  • Paul Fentress (2022; UC Berkeley; Now at Mentia)
  • Joseph Gawlik (2022; UC Berkeley)