Jihao Andreas Lin

Jihao Andreas Lin

Research Scientist at Meta

New York, NY, USA

About

I am currently a Research Scientist at Meta, working on optimization, hyperparameter scaling, and training stability for frontier AI model pre-training. Previously, I worked on probabilistic modeling and neural network training efficiency. Before joining industry, I completed my PhD at the University of Cambridge, where I studied probabilistic machine learning in the Computational and Biological Learning Lab under the supervision of José Miguel Hernández-Lobato.

Experience

  1. Present

    Meta

    Research Scientist

    AI & Systems Co-Design (with TBD Labs)
    • Optimization, hyperparameter scaling, and training stability for large language model pre-training.
    Central Applied Science
    • Probabilistic modeling for more efficient neural network training, delivering $48M/year in compute resource savings.
  2. 2025

    University of Cambridge

    Ph.D. in Probabilistic Machine Learning

    • Thesis: "Scalable Gaussian Processes: Iterative Methods and Pathwise Conditioning", supervised by José Miguel Hernández-Lobato
    Google · Student Researcher
    • Researched learning curve prediction for early stopping and model scaling.
    Meta · Research Scientist Intern
    • Developed learning curve prediction models for early stopping, demonstrating up to 25% compute savings on internal production data.
  3. 2022

    Smiths Detection

    Research Engineer

    • Used convolutional neural networks to enhance the quality of 3D CT baggage scans.
  4. 2021

    TU Darmstadt

    M.Sc. in Autonomous Systems

    • Thesis: "Functional Variational Inference in Bayesian Neural Networks", supervised by Jan Peters
    Visiting Student
    • Studied at University of British Columbia, Stanford University, and Aalto University.
    2018

    B.Sc. in Computer Science

    • Thesis: "Liquid Segmentation in 3D CT Baggage Scans", supervised by Stefan Roth
    Research Assistant
    • Research on convolutional neural networks and contributions to the MOT16 tracking benchmark.

Selected Publications

Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search

Z. Chen, S. Ament, D. Eriksson, M. Balandat, B. K. H. Low, E. Bakshy, J. A. Lin

Preprint

Empirical Gaussian Processes

J. A. Lin*, S. Ament*, L. C. Tiao, D. Eriksson, M. Balandat, E. Bakshy

International Conference on Machine Learning 2026

Scalable Gaussian Processes with Latent Kronecker Structure

J. A. Lin, S. Ament, M. Balandat, D. Eriksson, J. M. Hernández-Lobato, E. Bakshy

International Conference on Machine Learning 2025

Sampling from Gaussian Process Posteriors using Stochastic Gradient Descent

J. A. Lin*, J. Antorán*, S. Padhy*, D. Janz, J. M. Hernández-Lobato, A. Terenin

Advances in Neural Information Processing Systems 2023 (Oral Presentation)

Latent Derivative Bayesian Last Layer Networks

J. Watson*, J. A. Lin*, P. Klink, J. Pajarinen, J. Peters

International Conference on Artificial Intelligence and Statistics 2021

See Google Scholar for a complete list of publications.