Modular Norm RandOpt: Population-Efficient Ensembling through Architecture-Aware Perturbations
Under review at ICLR 2027
Accepted at NeurReps @ NeurIPS 2026 (Extended Abstract Track)
Undergraduate student The University of Osaka
Research interests: AI safety; AI for science.
Under review at ICLR 2027
Accepted at NeurReps @ NeurIPS 2026 (Extended Abstract Track)
Investigating how post-training affects the safety and reliability of language models, with a focus on understanding and evaluating changes in model behavior.
Coming soon.