Yi Yang (杨益)

Ph.D. Student @ HDSI, UCSD

Interests: Causality

Hi! I am a first-year PhD student in the Halıcıoğlu Data Science Institute at UCSD, supervised by Prof. Biwei Huang. Prior to UCSD, I earned my Statistics Bachelor’s degree from the School of the Gifted Young, University of Science and Technology of China in June 2025. During my undergrad, I am passionate about applying ML algorithms with statistical intuition to gain insights into interdisciplinary areas, like language acquisition, the documentation and phylogeny of Sino-Tibetan Languages. Please check my CV (PDF) and the blog A Summary of Undergrad Research for more.

Research

I am dedicated to advancing our understanding of the world through a causal lens. I focus on discovering and embedding cause-effect relationships in complex, real-world environments across three key levels:

Theoretical Guarantees: General and Scalable Causal Discovery The central aim in many scientific fields is to elucidate cause-effect relationships. I work on developing general and scalable causal discovery algorithms with strong theoretical guarantees. A key focus is extending these algorithms to reliably handle large-scale scenarios in the presence of hidden confounders, missing values, distribution shifts, and selection bias.
Causal-Inspired Interpretability: Understanding Neural Networks and Human Brains Neural networks have shown the promise of general intelligence, like human brains, beyond simple memorization. My goal is to move beyond traditional association-based interpretability methods to perform post-analysis on core ML frameworks (e.g., transformers, diffusion models, and agentic systems) with a causal perspective, making them more white-box and controllable. Concurrently, I wish to explore the causal mechanisms of the human brain, aiming to unify insights from both fields.
Causality-Empowered AI By synthesizing theoretical guarantees with intuitions from causal interpretability, my long-term vision is to develop an infant-like architecture that can automatically discover latent causal features and relationships from data.

Miscellaneous

One of my biggest dreams is to create a full-length animated film, designed in the aesthetic of traditional Chinese calligraphy, painting, and literature. I hope to make it a reality someday🥹！

Playing badminton and running regularly.

I grew up surrounded by animals🥳. The Chinese Box Turtles and Border Collie are my favoriate.

Movies, poems, and novels always bring me something new amid the trivialities of daily life (my DouBan).

Chinese Calligraphy has been a part of my life since kindergarten. More Interestingly, I even got a chance to learn how to make a mixed-hair brush from scratch during my undergrad😄.

If anything interests you, I am happy to introduce them to you in detail!

Selected Publications [full list]

(*) denotes equal contribution

COLINGOral

Transformer-based Speech Model Learns Well as Infants and Encodes Abstractions through Exemplars in the Poverty of the Stimulus Environment

Yi Yang*, Yiming Wang*, and Jiahong Yuan

In Proceedings of the 31st International Conference on Computational Linguistics
Oral Presentation
PDF Code
EMNLPFindings

Automated Tone Transcription and Clustering with Tone2Vec

Yi Yang, Yiming Wang, and Jiahong Yuan

In Findings of the Association for Computational Linguistics: EMNLP, 2024.

PDF Code
TheWebConf

GIF: A General Graph Unlearning Strategy via Influence Function

Jiancan Wu*, Yi Yang*, Yuchun Qian, Yongduo Sui, Xiang Wang, and Xiangnan He

In the International World Wide Web Conference, 2023.

Abstract PDF Code

With the greater emphasis on privacy and security in our society, the problem of graph unlearning — revoking the influence of specific data on the trained GNN model, is drawing increasing attention. However, ranging from machine unlearning to recently emerged graph unlearning methods, existing efforts either resort to retraining paradigm, or perform approximate erasure that fails to consider the inter-dependency between connected neighbors or imposes constraints on GNN structure, therefore hard to achieve satisfying performance-complexity trade-offs. In this work, we explore the influence function tailored for graph unlearning, so as to improve the unlearning efficacy and efficiency for graph unlearning. We first present a unified problem formulation of diverse graph unlearning tasks w.r.t. node, edge, and feature. Then, we recognize the crux to the inability of traditional influence function for graph unlearning, and devise Graph Influence Function (GIF), a model-agnostic unlearning method that can efficiently and accurately estimate parameter changes in response to a 𝜖-mass perturbation in deleted data. The idea is to supplement the objective of the traditional influence function with an additional loss term of the influenced neighbors due to the structural dependency. Further deductions on the closed-form solution of parameter changes provide a better understanding of the unlearning mechanism. We conduct extensive experiments on four representative GNN models and three benchmark datasets to justify the superiority of GIF for diverse graph unlearning tasks in terms of unlearning efficacy, model utility, and unlearning efficiency.