PeopleofSuperintelligence

Researcher

Yusuke Hirota

Yusuke Hirota is a researcher in AI fairness whose work identifies and reduces societal biases in vision-language datasets and models. At Osaka University, he earned a B.S. in engineering, an M.S. in information science and a Ph.D. in computer science. His research examined bias in visual question answering and image captioning, and developed mitigation methods including LIBRA and SANER. Since April 2025, he is a research scientist at NVIDIA Research Taiwan, where he focuses on fairness in multimodal tasks.

Now at
NVIDIA
Badges

Timeline

  1. 2016–2022

    Degrees at Osaka University

    He studies at Osaka University from 2016 to 2020 for a B.S. in the School of Engineering, then from 2020 to 2022 for an M.S. in the School of Information Science and Technology.

  2. 2021

    A visual question-answering paper

    In 2021, he is first author of “Visual Question Answering with Textual Representations for Images” at ICCV CLVL 2021.

  3. 2022

    A PhD and two bias papers

    In April 2022, he begins a Ph.D. in Osaka University’s Graduate School of Information Science and Technology under Noa Garcia and Yuta Nakashima. His paper “Quantifying Societal Bias Amplification in Image Captioning” is an oral at CVPR 2022, and he coauthors “Gender and Racial Bias in Visual Question Answering Datasets” at ACM FAccT 2022.

  4. 2023

    LIBRA and Sony AI

    His CVPR 2023 paper with Yuta Nakashima and Noa Garcia introduces LIBRA, which learns from synthetically biased samples to reduce gender misclassification and replace stereotypical words in captions. He also coauthors “Uncurated Image-Text Datasets: Shedding Light on Demographic Bias,” selected as a CVPR 2023 highlight, and starts an internship at Sony AI in April 2023.

  5. 2023–2024

    NVIDIA internship and EMNLP papers

    In December 2023, he starts an internship at NVIDIA Research. He coauthors the CVPR 2024 paper “Would Deep Generative Models Amplify Bias in Future Models?” and two main-track EMNLP 2024 papers. “From Descriptive Richness to Bias: Unveiling the Dark Side of Generative Image Caption Enrichment” finds that enriched captions carry more gender bias and hallucination, and that models trained on them amplify gender bias by an average of 30.9%; “Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes” uses text-guided inpainting to make protected groups independent of all image attributes.

  6. 2025

    A doctorate and SANER

    His ICLR 2025 paper SANER, coauthored with Min-Hung Chen, Chien-Yi Wang, Yuta Nakashima, Yu-Chiang Frank Wang and Ryo Hachiuma, removes attribute information from CLIP text features of attribute-neutral descriptions without requiring attribute annotations. In March 2025, he completes a Ph.D. in computer science at Osaka University.

  7. 2025

    LOTUS and bias benchmarks

    In 2025, he is first author of LOTUS, an ACL industry-track leaderboard for detailed image captions that evaluates quality, hallucination and societal bias. His ICCV 2025 study finds that masking 10% of objects or weakly blurring backgrounds can shift gender-bias scores by up to 175% in generative vision-language models. In May 2025, he is selected as an outstanding reviewer for CVPR.

  8. Since April 2025

    Research scientist at NVIDIA

    In April 2025, he becomes a research scientist at NVIDIA Research Taiwan. In 2026, his papers are accepted at CVPR, ACL, ECCV and EMNLP, including “Guardrail-Agnostic Societal Bias Evaluation in Large Vision-Language Models” at ECCV 2026. At NVIDIA Research Taiwan, he focuses on fairness in multimodal tasks by identifying societal biases in datasets and models and developing mitigation strategies.

Career

  1. Since 2023NVIDIAIntern, Research Scientist
  2. From 2023SonyIntern

Education

  1. 2025Osaka UniversityPhD
  2. 2022Osaka UniversityMaster of Science
  3. 2020Osaka UniversityBachelor of Science

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Yusuke Hirota. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. Google has the highest combined h-index among AI organizationsIncludes NVIDIA. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes NVIDIA. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes NVIDIA. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Yusuke Hirota yet. Send us a free-licensed photo.

  1. Yusuke Hirota y-hirota.com. Checked 22 October 2026.
  2. Yusuke Hirota | NVIDIA Research Taiwan research.nvidia.com. Checked 22 October 2026.
  3. Yusuke Hirota | ISLab, Osaka University is.ids.osaka-u.ac.jp. Checked 22 October 2026.
  4. Model-agnostic gender debiased image captioning | ISLab, the University of Osaka is.d3c.osaka-u.ac.jp. Checked 22 October 2026.
  5. [2406.13912] From Descriptive Richness to Bias: Unveiling the Dark Side of Generative Image Caption Enrichment arxiv.org. Checked 22 October 2026.
  6. [2407.03623] Resampled Datasets Are Not Enough: Mitigating Societal Bias Beyond Single Attributes arxiv.org. Checked 22 October 2026.
  7. [2408.10202] SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP arxiv.org. Checked 22 October 2026.
  8. [2507.19362] LOTUS: A Leaderboard for Detailed Image Captioning from Quality to Societal Bias and User Preferences arxiv.org. Checked 22 October 2026.
  9. [2509.07596] Bias in Gender Bias Benchmarks: How Spurious Features Distort Evaluation arxiv.org. Checked 22 October 2026.