About Me

I am an MSc student in Computer Science at the University of São Paulo (IME-USP), advised by Prof. Nina S. T. Hirata. My research lies at the intersection of Large Language Models, Geometric Deep Learning, and AI for Science. My research is funded by FAPESP.

Previously, I was a research intern at IBM Research, working on geometric deep learning for materials science with Dr. Mathias Steiner’s team. I was also a research intern at the University of Texas at Austin, hosted by Prof. Zhangyang “Atlas” Wang (VITA Group), where I worked closely with Prof. Junyuan “Jason” Hong on Large Language Models. I received my BSc in Computer Science from IME-USP.

CV GitHub LinkedIn Scholar ORCID Lattes Email

News

• 2026-01 AD-VF accepted at ICRA 2026.
• 2026-01 Geodite preprint is out.
• 2026-01 Started my MSc in Computer Science at the University of São Paulo (IME-USP).

Selected Publications

• Equivariant Interatomic Potentials without Tensor Products.
  T. Reschützegger, S. Aykent, Gabriel Jacob Perin, B. H. Nunes, F. Cipcigan, R. N. B. Ferreira, M. Steiner, F. L. Thiemann.
  Preprint 2026 [Paper] [Code] [Model]
• LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning.
  Gabriel Jacob Perin, R. Chen, X. Chen, N. S. T. Hirata, Z. Wang, J. Hong.
  COLM 2025 [Paper] [Code]
• RankMean: Module-Level Importance Score for Merging Fine-tuned LLM Models.
  Gabriel Jacob Perin, X. Chen, S. Liu, B. Kailkhura, Z. Wang, B. Gallagher.
  ACL 2024 (Findings) [Paper] [Code]
• Few-shot Retinal Disease Classification on the Brazilian Multilabel Ophthalmological Dataset.
  Gabriel Jacob Perin, N. S. T. Hirata.
  SIBGRAPI 2024 [Paper] [Code]

Other

• Undergraduate Thesis.
• MAC5856 Blog Posts.
• Languages: Portuguese (native), English (fluent), Spanish (basic).