About

Hi! I am an incoming postdoc at Stanford working with Prof. Tatsu Hashimoto and Prof. Percy Liang. Recently, I finished my PhD at MIT, where I was fortunate to be advised by Prof. Aleksander Mądry.

I’m currently interested in understanding and improving machine learning (ML) methodology through the lens of data. Some questions I think about include:

• How do we attribute model predictions back to training data?
• How do we select the right data for a given task?
• Can we derive insights about ML phenomena (e.g., scaling laws, emergence, in-context learning) through this lens?

I’m also more broadly interested in the science of machine learning/deep learning.

[News] I co-presented a tutorial at ICML ‘24 on Data Attribution at Scale: [video] [notes]!

 

Bio

Previously at MIT, I worked on understanding statistical-computational tradeoffs in high-dimensional statistics with Prof. Guy Bresler for my SM thesis. Earlier during my PhD, I was supported by the MIT Akamai Presidential Fellowship and the Samsung Scholarship.

From 2016-18, I served in the Republic of Korea Army in the top signals intelligence unit as a researcher.

Prior to grad school, I received a BS in Computer Science from Cornell University (2011-14), where I was fortunate to work with Prof. Ramin Zabih and Prof. Bobby Kleinberg.

I have interned at Waymo, Dropbox, and Google.

 

Research

The Journey, Not the Destination: How Data Guides Diffusion Models
Kristian Georgiev*, Josh Vendrow*, Hadi Salman, Sung Min Park, Aleksander Mądry
[arxiv]

TRAK: Attributing Model Behavior at Scale
Sung Min Park*, Kristian Georgiev*, Andrew Ilyas*, Guillaume Leclerc, Aleksander Mądry
ICML 2023 (Oral presentation)
[arxiv] [blog][code] [website][talk]

ModelDiff: A Framework for Comparing Learning Algorithms
Harshay Shah*, Sung Min Park*, Andrew Ilyas*, Aleksander Mądry
ICML 2023
[arxiv] [blog][code]

FFCV: Accelerating Training by Removing Data Bottlenecks
Guillaume Leclerc, Andrew Ilyas, Logan Engstrom, Sung Min Park, Hadi Salman, Aleksander Mądry
CVPR 2023
[code]

A Data-Based Perspective on Transfer Learning
Saachi Jain*, Hadi Salman*, Alaa Khaddaj*, Eric Wong, Sung Min Park, Aleksander Mądry
CVPR 2023
[arxiv] [blog]

Datamodels: Predicting Predictions from Training Data
Andrew Ilyas*, Sung Min Park*, Logan Engstrom*, Guillaume Leclerc, Aleksander Mądry
ICML 2022
[arxiv] [blog part 1 part 2] [code][data]

On Distinctive Properties of Universal Perturbations
Sung Min Park, Kuo-An Wei, Kai Xiao, Jerry Li, Aleksander Mądry
2021
[arxiv]

Sparse PCA from Sparse Linear Regression
(α-β order) Guy Bresler, Sung Min Park, Madalina Persu
NeurIPS 2018
[arxiv] [poster] [code]

On the Equivalence of Sparse Statistical Problems
Sung Min Park
SM thesis 2016
[pdf]

Structured learning of sum-of-submodular higher order energy functions
Alexander Fix, Thorsten Joachims, Sung Min Park, Ramin Zabih
ICCV 2013
[pdf]

 

Talks

• Mar 2024 Stanford ML lunch
• Jul 2023 ICML Oral
• May 2023 LIDS & Stats Tea
• May 2023 MIT MLTea
• Apr 2023 ML Collective Reading group
• Feb 2023 MIT LIDS Student Conference
• Aug 2022 UMN ML Seminar
• Feb 2022 LIDS & Stats Tea
• Jan 2022 MIT LIDS Student Conference

 

Misc

Region Detection and Geometry Prediction
Patent from work during Summer 2020 internship at Waymo
[pdf]

Fourier Theoretic Probabilistic Inference over Permutations
Cornell, Spring 2014
[pdf]

Analysis of pipage method for k-max coverage
Cornell, Fall 2012
[pdf]

 

Personal

I grew up between the Bay Area, Seoul, and Singapore, where I attended SAS.

In my free time, I enjoy lifting, playing basketball, rowing, watching the NBA (nuggets!), watching movies, and learning physics and math.