Papers by Bryan R Christ
EDUMATH: Generating Standards-aligned Educational Math Word Problems (2026.acl-long)
Copied to clipboard
Bryan R Christ, Penelope Molitz, Beau LeBlond, Zachary Gottesman, Jonathan Kropko, Thomas Hartvigsen
| Challenge: | Math word problems (MWPs) are critical elements of K-12 math education and can be customized to students' interests and ability levels. |
| Approach: | They propose that LLMs can generate MWPs customized to student interests and math education standards by using an open and closed LLM to evaluate over 11,000 MWps and develop a teacher-annotated dataset for standards-aligned educational MWPS generation. |
| Outcome: | The proposed model outperforms existing closed models without training and is more similar to human-written MWPs but prefers customized MWPS with grade school students. |
Math Neurosurgery: Isolating Language Models’ Math Reasoning Abilities Using Only Forward Passes (2025.acl-long)
Copied to clipboard
| Challenge: | Math reasoning is an active area of Large Language Model (LLM) research because it is a hallmark of artificial intelligence and has implications in several domains, including math education. |
| Approach: | They propose a method to isolate math-specific parameters in LLMs using only forward passes. |
| Outcome: | The proposed method improves a model's performance on GSM8K and MATH by 4-17% while leaving non-math behavior unaltered. |