Eric Xue

I'm Eric Xue, a fourth-year undergraduate student studying computer science at the University of Toronto. I use technology to create innovations that open up new possibilities and lead us toward more exciting futures.

I'm currently a student researcher at the

Email  /  Resume  /  Twitter  /  Github

profile photo

Research

I have explored a range of research topics such as large language models, adversarial robustness, dataset distillation, computer vision, and reinforcement learning. Recently, I am mostly interested in studying the potential applications of LLMs.

Towards Adversarially Robust Dataset Distillation by Curvature Regularization
Eric Xue, Yijiang Li, Haoyang Liu, Yifan Shen, Haohan Wang
Accepted to AAAI 2025

Proposed a curvature regularization technique, informed by theoretical insights, to enhance the inherent adversarial robustness of models trained on condensed datasets.

AutoModel: Autonomous Model Development for Image Classification with LLM Agents
Eric Xue, Zeyi Huang, Yuyang Ji, Haohan Wang
Under Review at ICLR 2025

An LLM agent framework that builds and optimizes image classification models by simulating a human machine learning development team. AutoModel iteratively improves all aspects of the pipeline, including data augmentation, model architecture, optimizers, and hyperparameters, from real training feedback and entirely without human supervision.

Learning to Imitate with Less: Efficient Individual Behavior Modeling in Chess
Zhenwei Tang, Difan Jiao, Eric Xue, Reid McIlroy-Young, Jon Kleinberg, Siddhartha Sen, Ashton Anderson
Under Review at ICLR 2025

A personalized AI behavior model for chess that uses two-stage fine-tuning and meta-networks to efficiently predict individual decision-making styles with minimal data, enabling flexible human-AI collaboration.

Towards Machine Theory of Mind with Large Language Model-Augmented Inverse Planning
Rebekah A. GelpĂ­, Eric Xue, William A Cunningham
Under Review at ICLR 2025

A hybrid machine Theory of Mind approach combining LLM-generated hypotheses with Bayesian inverse planning to predict mental states, enhancing performance on ToM tasks and enabling socially intelligent agent models.

Granular Analysis of Pretrained Object Detectors
Eric Xue, Tae Soo Kim
Published at ICAIIC 2022

Granular performance analysis using ROC curves of pre-trained object detectors in the autonomous vehicle setting, examining different data subgroups (bounding box sizes, object types, occlusion levels) and various image perturbations.

Selected Projects

Mirai

An LLM-based application essay editing platform built with the MERN stack that rewrites and enhances student essays, highlighting their strengths and aligning with their interests, activities, desired schools, and intended major for undergraduate or graduate applications, aiming to provide equal opportunities and guidance for all students in the college application process.

Yumo Bot

An AI clone of myself, fine-tuned using scraped text data from my Discord conversations, deployed as a bot on Discord to explore the integration of AI/LLMs into everyday life.

Online AI Character Chat

My own version of an AI character chatting platform that incorporates features I found missing in similar platforms, such as Character AI, including avatar generation for AI characters and autonomous messaging from AI characters to users.

Buddy Breed

An iOS app, previously published on the App Store (now removed due to lack of updates), that allowed users to identify dog breeds from a single photo by utilizing fine-tuned image classification models in the backend.

Miscellanea

Awards

Woodsworth College Scholarship - 2024
Dean's List Scholar - 2022, 2023, 2024
BC Achievement Scholarship - 2021
BC District/Authority Scholarship - 2021
Google Code-in Runner Up - 2019
RoboCupJunior Soccer Worlds - 2nd Place 2019; 3rd Place 2017