Yuchong Geng

I am a Ph.D. candidate in Electrical and Computer Engineering at Cornell University, advised by Prof. A. Kevin Tang. My research explores cognitively inspired learning systems that discover and reason with conceptual knowledge, aiming to align machine learning with human cognition processes.

Email / Google Scholar / LinkedIn / GitHub

News

06/2026: Atelier, a side project of ours on agentic AI research workflows, has been accepted to the ICML 2026 Workshop AI4Research.
01/2026: Our paper has been accepted to TMLR.
12/2025: Completed my internship at Qualcomm in San Diego 🌴.
07/2025: Passed my A Exam and advanced to PhD candidacy at Cornell ECE.
07/2022: I will continue my education at Cornell as a PhD student in the School of ECE.
11/2021: I gave a talk at CoRL 2021 in London, UK.
10/2021: A paper about my research work during my MEng study at Cornell has been accepted to CoRL 2021.

Education

Ph.D. in Electrical and Computer Engineering, Cornell University
M.Eng. in Electrical and Computer Engineering, Cornell University
B.S. in Electrical Engineering, University of California, Davis

Industry Experience

Incoming Intern, Google — Seattle, WA (Fall 2026)
PhD Software Engineer Intern, Meta — Menlo Park, CA (Summer 2026)
Research Intern, Qualcomm — San Diego, CA (Fall 2025)

Research

My research focuses on building concept-centric and cognitively grounded learning systems that aim to bridge modern machine learning with human cognition. My goal is to design learning frameworks that can discover patterns from data, organize them into persistent conceptual knowledge, and reuse this knowledge to generalize across tasks and modalities.

	Unsupervised Concept Discovery with Dirichlet Concept Diffusion Models Under double-blind review project page DCDM learns reusable visual concept centers while training a diffusion model, using an input-dependent Dirichlet distribution to compose concepts without labels, captions, or semantic concept supervision.
	Atelier: A Human-Owned Protocol for Auditable AI-Assisted Research Accepted to ICML 2026 Workshop AI4Research project page Atelier is a file-backed protocol for AI-assisted research that keeps the researcher in charge while routing agent work through scoped manager, worker, and critic roles with auditable evidence.
	A Concept-Centric Approach to Multi-Modality Learning Yuchong Geng, Ao Tang Transactions on Machine Learning Research (TMLR), 2026 arXiv / project page Humans reuse knowledge across modalities and tasks. This project builds ML systems that do the same by learning reusable concepts in a shared space and connecting multiple modalities through lightweight projection modules.
	Decentralized Sharing and Valuation of Fleet Robotic Data Yuchong Geng, Dongyue Zhang, Po-han Li, Oguzhan Akcin, Ao Tang, Sandeep P. Chinchali CoRL, 2022 paper / project page Robots operating in different environments learn to identify, value, and exchange rare but important experiences without centralized data aggregation.

Teaching

ECE 3100: Introduction to Probability and Inference for Random Signals and Systems
ECE 4450/5660: Computer Networks and Telecommunications

Template adapted from the Jon Barron's website. Source repo: jonbarron/jonbarron.github.io.