Welcome

My name is Yihong Chen. I study AI knowledge acquisition, with a particular focus on how systems learn to abstract, represent, and use concepts/symbols efficiently.

News

• Jul 2025, My PhD thesis is released: Knowledge engines need not just structure, but also destructuring — for plasticity, flow, and adaptability.
• Mar 2025, We introduced a new tool for transformer interpretability: jet expansion.
• March 2024, Quanta Magazine covered our research on active forgetting. The article is available here.
• Dec 2023, Presentation of our work on forgetting at NeurIPS 2023 (poster available here).
• Sep 2023, Talk at the IST–Unbabel Seminar on learning with forgetting.
• Jul 2023, Presentation at ELLIS Unconference 2023 on forgetting in language modeling (slides available here).
• Jul 2023, Our paper Improving Language Plasticity via Pretraining with Active Forgetting demonstrates how forgetting-enhanced pretraining enables rapid adaptation to new languages.
• Nov 2022, REFACTOR GNNs: Revisiting Factorisation-based Models from a Message-Passing Perspective appeared at NeurIPS 2022.
• Jun 2022, We released ssl-relation-prediction, a hands-on repository for experimenting with link prediction.