|
I am a MSc student in Computer Science and Engineering at the University of California, Santa Cruz. I am also a researcher at Professor Jeffrey Flanigan's lab, a place where I have enjoyed working for the last two and a half years. I will apply for PhD in Fall 2022. My research interests broadly lie in {deep learning} ∪ {computational neuroscience}. My long-term research goal is to leverage knowledge from both machine learning and the brain to build intelligent systems that can learn adaptively, continuously, and with interpretability.My current research focuses on fundamental problems in deep learning. I am particularly interested in demystifying generalization properties and behaviors in large and small neural networks (e.g., double descent, expressive capacity of large networks, data and model scaling laws, learning noise, lottery tickets, out-of-distribution generalization, neural tangent kernel, etc.). I am also working with Professor Yang Liu on unifying bias in data and machine. Previously, I obtained my B.S. in Computer Science and Engineering at UC Santa Cruz. My last name, Liu (刘 or 劉), is pronounced "Leo" but with shorter "[I]" sound. |
|
Previous events
|
|
These included publications and preprints. |
|
|
We empirically estimate the power-law exponents of various model architectures and study how they are altered by a wide range of training conditions for classification. |
|
We present a dataset filtering approach that uses sets of classifiers, similar to ensembling, to estimate noisy (or non-realizable) examples and exclude them so a faster sample complexity rate is achievable in practice. |
|
These include coursework and side projects. |
|
We study the relationship between the amount of mutual information compression and generalization given the double descent phenomenon. |
|
We study the problem of learning to construct compact representations of neural network weight matrices by projecting them into a smaller space. |
|
We surveyed recent works on dataset bias and machine learning bias. |
|
We show that adversarially training (Fast Gradient Sign Method and Projected Gradient Descent) reduces the empirically sample complexity rate for MLP and a variety of CNN architectures on MNIST and CIFAR-10. |
|
Trained distilled RoBERTa model as a text classifier and a GPT-2 as a text generator trained using proximal policy optimization synchronously to generate augmented text for text classification tasks. |
|
Fine-tuned a GPT-2 model using all research paper titles and abstracts under cs.AI, cs.LG, cs.CL, and cs.CV on arXiv. This project was the winner of the Generative Modeling Competition for the course CSE142 Machine Learning in Spring 2020. |
|
Fine-tuned a RoBERTa model on the IMDb dataset for sentiment analysis. This project was the winner of the Sentiment Analysis Competition for the course CSE142 Machine Learning in Spring 2020. |
|
|
This is a fork of Jon Barron"s website. |