Gao, Yansong (髙岩🌲)

alt text 

PhD Candidate,
GRASP Robotics Laboratory,
University of Pennsylvania
Philadelphia, USA
E-mail: gaoyans@sas.upenn.edu

About me

Hi! I am a Ph.D. Candidate at the University of Pennsylvania and advised by Pratik Chaudhari. I work on a handbook for pre-training models. My research interests span areas like pre-training, transfer learning, unsupervised learning, generative models, and information theory. I graduated from Shanghai Jiao Tong University (SJTU) in 2017 with a Dual Degree (Bachelor in Mathematics and Physics), where Yaokun Wu advised me.

Research

My research interests include:

  • Multi-task Learning, Transfer Learning, and Unsupervised Learning

  • Deep Gnerative Models, Foundation Models

  • Information Geometry, Information Theory

  • Algorithmic game theory

Current work

  • A handbook for pretraining learning models

Selected Publications

  • Fast Diffusion Probabilistic Model Sampling through the lens of Backward Error Analysis

    Yansong Gao*, Zhihong Pan, Xin Zhou, Le Kang, and Pratik Chaudhari

    [PDF]

    In submission 2023

  • Deep Reference Priors: What is the best way to pretrain a model?

    Yansong Gao*, Rahul Ramesh*, and Pratik Chaudhari

    ICML 2022

    [ArXiv][code]

  • Beyond the worst-case analysis of random priority: Smoothed and average-case approximation ratios in mechanism design

    Xiaotie Deng, Yansong Gao, and Jie Zhang

    Information and Computation 2022

    [PDF]

  • An information-geometric distance on the space of tasks

    Yansong Gao and Pratik Chaudhari

    ICML 2021

    [PDF][code]

  • A free-energy principle for representation learning

    Yansong Gao and Pratik Chaudhari

    ICML 2020

    [PDF][code]

  • Average-case analysis of the assignment problem with independent preferences

    Yansong Gao and Jie Zhang

    IJCAI 2019

    [PDF]

  • On Scheduling Mechanisms Beyond the Worst Case

    Yansong Gao and Jie Zhang

    [ArXiv]

  • Full list of publications in Google Scholar.

Reviewer

  • ICML2020, ICML2021, ICML2022, NeurIPS2021, NeurIPS2022, ICLR2022, Entropy

Puzzles