Jiwon's Alcove
Research Blog

Submodular coresets: Machine learning with less data

Feb 28, 2025

A theoretical computer science problem that the SOTA LLMs get wrong (as of now)

Feb 21, 2025

A P-splinter is a language where a PTIME function can enumerate its elements. Despite this being an old result, state-of-the-art LLMs get this problem wrong. (For now—by writing this, I am inadvertently helping the next generation of LLMs improve via memorization.)

CoolerSpace: A Language for Physically Correct and Computationally Efficient Color Programming

2024

The PLUTUS pipeline consists of a cycle: model training, sliceline to identify problematic subgroups, and distribution tailoring to acquire additional data.

PLUTUS: Understanding Data Distribution Tailoring for Machine Learning

2024

The Two Coupons, Generic Quota Coupon Collector's Problem

Jul 7, 2023

Suppose there are two distinct coupons. In each iteration, the probability of sampling type-1 coupon is $p$ and that for type-2 coupon is $q = 1 - p$. Our goal is to collect at least $k$ of type-1 and $r$ of type-2. How many iterations does it take, in expectation, to complete a full collection?
The DT pipeline where data sources are combined with RatioColl or EpsilonGreedy to form a balanced unified dataset.

Data Distribution Tailoring

2022-2024

Red and blue points on a plane, with some red points and some blue points having a circle around them.

Fair $k$-Cover Coresets

2022

The pipeline. An SDR image is made brighter and darker, then denoising is applied. The resultant three images are combined with Mertens' fusion.

Single Exposure Fusion

2021