Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
Understanding LLM Reasoning Through Meaning-Removed Steering Vectors
Published:
Large Language Models (LLMs) have shown remarkable capabilities in reasoning tasks, but understanding and controlling their internal reasoning processes remains a significant challenge. In my ongoing research at the MINE Lab (University of Notre Dame), I’m working on a novel approach to this problem: meaning-removed steering vectors.
portfolio
Benchmark Harmonization and Model Similarity Analysis
Harmonizing major AI evaluation benchmarks (LiveBench, HELM, LMSYS Arena) and developing model similarity maps at FORESEER Lab, University of Michigan.
Meaning-Removed Steering Vectors for LLM Reasoning
Developing novel techniques for calibrating large language model reasoning through sentence-level hidden-state interventions at MINE Lab, University of Notre Dame.
Trajectory-Level Web Agent Evaluation
Developing comprehensive evaluation frameworks for web agents that assess both action sequences and value alignment at SaNDwich Lab (IBM–Notre Dame collaboration).
ML Verification Benchmarks and Reproducibility
Improving machine learning verification benchmarks and addressing numerical reproducibility challenges with the alpha-beta-CROWN verification tool at UIUC.