Publications
Lightweight Learner for Shared Knowledge Lifelong Learning
Yunhao Ge, Yuecheng Li*, Di Wu*, Ao Xu*, Adam M. Jones, Amanda Sofie Rios, Iordanis Fostiropoulos, Shixian wen, Po-Hsuan Huang, Zachary William Murdock, Gozde Sahin, Shuo Ni, Kiran Lekkala, Sumedh Anand Sontakke, Laurent Itti (*=equal contribution as second author)
TMLR (Transactions on Machine Learning Research)
Research Experience
- Research Assistant, CNRL [Supervised by Prof. Nicolas Schweighofer], University of Southern California, 2023-Present Keywords: AI for Science, Bayesian Inference, Reinforcement Learning
- Developed reinforcement learning model for human motor learning and used Bayesian inference to estimate the parameters.
- Research Assistant, iLab [Supervised by Prof.Laurent Itti], University of Southern California, 2021-Present Keywords: Transfer Learning, Continual Learning
- Created algorithm for transfer and continual learning with SOTA performance (86.25% (ours) vs. 81.53% (Head2Toe) in transfer learning and vs. 56.69% (Supermasks in Superposition) in continual learning) and low extra memory (only 0.59% extra parameters needed)
- Defined a multi-agent life-long learning task, created a dataset (>100 tasks) for the problem, and achieved SOTA performance compare to other algorithms such as GEM and derpp. (60% (ours) compare to 32%) with lower memory cost and less training time
- Created three websites for publications in lifelong learning on different projects.
- Research Assistant, ICAROS [Supervised by Prof.Stefanos Nikolaidis], University of Southern California, 2021-2022 Keywords: Quality Diversity Algorithm, Optimization
- Applied Differentiable Quality Diversity algorithm to generate numerous valid solutions within a given measurement
- Generated life-like digits images for the MNIST generation task with various likelihoods of digits other than themselves
- Research Assistant, ISI [Supervised by Prof.Emilio Ferrara], University of Southern California, 2021-2022 Keywords: Natural Language Process
- Created an algorithm to identify the meaning of newly emerging words on the Internet with FastText
- Utilized nltk and FastText for data preprocessing and word encoding respectively
- Research Assistant, Intelligent Motion Laboratory [Supervised by Prof.Kris Hauser], University of Illinois - Urbana Champaign, 2019-2021 Keywords: Robotic
- Created an API to control the gripper of a nursing robot to complete pick, grab and deliver tasks. Created the program for 3D object segmentation using clustering for the robot to “see” the object. Created algorithm to generate grasping policy with DQCNN for robots to successfully pick the target
- Earned a finalist position in ANA Avatar XPRIZE Competition
Work Experience
- Research Intern, Anlogic Co. Ltd., 2019-2019
- Developed algorithms using machine learning and statistical modeling techniques to increase performance, quality, data management, and accuracy
- Optimized the company's product - segmentation network's size by 10% using pruning and quantization technique
- Decreased time and memory utilization without a statistically significant decrease in accuracy