Research

My research focuses on trustworthy NLP, AI safety, LLM interpretability, and representation analysis.

I am particularly interested in understanding and mitigating subtle biases in language and multimodal models, while improving their reliability and generalization in real-world settings.

My recent work spans embedding-level bias mitigation, bias evaluation in vision-language models, and post-training methods for preserving generalization in vision-language-action systems.

Current Research

• OMRON SINIC X Corporation, Japan
  Full-time Research Intern (Jan 2026 - Present)
  Working on preserving generalization performance during VLA post-training.

• Ochanomizu University, Japan
  M.S. Researcher in Computer Science (Apr 2025 - Present)
  Studying embedding-level bias mitigation by controlling output dimensions associated with token probabilities.

Previous Research

• Ochanomizu University, Japan
  B.S. Researcher in Computer Science (Apr 2024 - Mar 2025)
  Developed a DFT-based approach on log-likelihood sequences to detect subtle gender bias beyond probability-gap analyses.

• Ochanomizu University, Japan
  Pre-Graduation Undergraduate Research Project (Apr 2023 - Mar 2024)
  Analyzed gender-gap factors across 47 prefectures in Japan using large-scale data collection and visualization.

Visiting / Industry Research Experience

• DFKI Saarbruecken, Germany
  Visiting Researcher (Aug 2025 - Oct 2025)
  Conducted collaborative research on trustworthy and interpretable AI.

• vislang Lab, Rice University, U.S.
  Visiting Student (Feb 2025 - Aug 2025)
  Built a human-centered benchmark and evaluated 39 VLMs for reasoning, bias, and identity verification.

• Artificial Intelligence Research Center at Fujitsu Laboratories Ltd., Japan
  Short-term Research Intern (Aug 2024)
  Developed prompt-based personality conditioning and evaluation metrics for bias detection in interview-assistant settings.