Joseph Suh

I'm a second year Ph.D. student at BAIR in UC Berkeley, co-advised by Professor Serina Chang and Professor John Canny.

Email / CV / Scholar / X/Twitter / Github

Research

I'm interested in deep learning and generative models. Most of my recent research is about language models as models of human behavior, spanning around opinions, actions, preferences, and temporal changes. Some papers are highlighted.

	Higher-Order Binding of Language Model Virtual Personas: a Study on Approximating Political Partisan Misperceptions Minwoo Kang, Suhong Moon, Seung Hyeong Lee, Ayush Raj, Joseph Suh, David M. Chan Preprint, 2025 arXiv / Code Expanding evaluations of LLM virtual personas to include ingroup / outgroup and meta-perception, "backstories" serve as a stepping stone to higher-order reflections in social context.
	Language Model Fine-Tuning on Scaled Survey Data for Predicting Distributions of Public Opinions Joseph Suh, Erfan Jahanparast, Suhong Moon, Minwoo Kang, Serina Chang Preprint, 2025 arXiv / Code / Dataset (SubPOP) Fine-tuning LLMs on response distributions from public opinion survey questions enables the models to predict opinions across different subpopulations, survey waves, and survey families.
	Rediscovering the Latent Dimensions of Personality with Large Language Models as Trait Descriptors Joseph Suh, Suhong Moon, Minwoo Kang, David M. Chan NeurIPS Workshop on Behavioral ML*, 2024 arXiv Constructing latent dimensions of personality using log-probabilities from language models, inspired by the methodology psychologists used to develop the Big Five model.
	Virtual Personas for Language Models via an Anthology of Backstories Suhong Moon, Minwoo Kang, Marwa Abdulhai, Joseph Suh*, Widyadewi Soedarmadji, Eran Kohen Behar, David M. Chan, John Canny EMNLP*, 2024 arXiv / Code We propose naturalistic bodies of text describing individual life stories, namely "backstories", as prefix to model prompts for persona conditioning.
	Long-range-interacting topological photonic lattices breaking channel-bandwidth limit Gyunghun Kim, Joseph Suh, Dayeong Lee, Namkyoo Park, Sunkyu Yu Nature Light: Science & Applications, 2024 Journal link A photonic lattice with non-nearest-neighbor interactions that breaks the tradeoff between topological channels and channel bandwidths.
	Photonic topological spin pump in synthetic frequency dimensions Joseph Suh, Gyunghun Kim, Hyungchul Park, Shanhui Fan, Namkyoo Park, Sunkyu Yu Physical Review Letters, 2024 Journal link Theoretical modeling of Laughlin's topological pump in synthetic frequency dimensions by the interplay of frequency mode-dependent and independent gauge fields.

Education

University of California, Berkeley
Ph.D. program in Electrical Engineering and Computer Sciences
Berkeley Artificial Intelligence Research Lab (BAIR)
Advisors: Prof. Serina Chang and Prof. John Canny
09/2023 – Present

Seoul National University
B.S. in Electrical & Computer Engineering
03/2017 – 03/2023

Gyeonggi Science High Schhol
03/2014 – 03/2017

About me

Before joining to Berkeley, I graduated from Seoul National University with a B.S. in Electrical and Computer Engineering, although I spent more time in the Physics building. In my undergrad years, I was honored to work with Prof. Sunkyu Yu in the field of photonics and condensed matter physics. Before my undergrad, I was a big fan of competitive programming, inspired by my close friends—some of whom continue pursuing their passions in theoretical computer science.

Honors and Awards

KFAS Doctoral Study Abroad Scholarship | 2023
Around 40 students selected nationally

Presidential Science Scholarship | 2017

Blog

CUDA kernel fusion
Lessons learned from implementing back-to-back GEMM kernel for LoRA serving in vLLM inference engine.
Given a skinny matrix A and a fat matrix B, how can we make xAB into a single kernel?

CS280A portfolio
Assignments and projects. Pretty much enjoyed hands-on experience, like warping, diffusion, GANs.