PeopleofSuperintelligence

Researcher

Golnaz Ghiasi

Golnaz Ghiasi is a researcher at Google DeepMind whose work runs from object detection to olympiad mathematics. Advised by Charless Fowlkes, they completed a PhD at the University of California, Irvine in 2016 on recognizing and segmenting objects under occlusion and clutter. With Tsung-Yi Lin and Quoc Le, they coauthored DropBlock in 2018 and NAS-FPN in 2019, and their Simple Copy-Paste and OpenSeg papers of 2021 and 2022 took on instance segmentation and open-vocabulary segmentation. In July 2025, an advanced version of Gemini Deep Think reached a gold-medal score at the International Mathematical Olympiad, with Ghiasi among the leads and key contributors of the team. Their work on mathematical reasoning continued with the IMO-Bench benchmarks in 2025 and, in February 2026, the paper introducing Aletheia, a math research agent.

Timeline

  1. 2015–2016

    A PhD at UC Irvine

    Advised by Charless Fowlkes, they earn a PhD at the University of California, Irvine in 2016 with the dissertation "Recognizing and Segmenting Objects in the Presence of Occlusion and Clutter". Their joint papers with Fowlkes include a 2015 face detector that explicitly models occluded parts and a 2016 Laplacian pyramid architecture for semantic segmentation.

  2. 2017

    Real-time arbitrary style transfer

    In 2017, they present a neural artistic stylization network at the British Machine Vision Conference with Honglak Lee, Manjunath Kudlur, Vincent Dumoulin and Jonathon Shlens. Trained on roughly 80,000 paintings, the model stylizes images in real time using any content and style image pair.

  3. 2018–2019

    DropBlock and NAS-FPN

    With Tsung-Yi Lin and Quoc Le, they introduce DropBlock at NeurIPS in 2018, a structured form of dropout that drops contiguous regions of a feature map together. NAS-FPN, accepted at CVPR in 2019 with the same coauthors and Ruoming Pang, uses neural architecture search to discover a feature pyramid architecture for object detection.

  4. 2021

    Simple Copy-Paste augmentation

    A CVPR 2021 paper they write with Yin Cui, Aravind Srinivas, Barret Zoph and others shows that randomly pasting objects onto images is a strong data augmentation for instance segmentation. On COCO, the method reaches 49.1 mask AP.

  5. 2022

    OpenSeg for open-vocabulary segmentation

    OpenSeg, accepted at ECCV 2022 with Xiuye Gu, Yin Cui and Tsung-Yi Lin, organizes an image into regions named by arbitrary text. The model learns from image-level captions by aligning each word in a caption to one or a few predicted masks.

  6. December 2023

    The Gemini model family

    They coauthor the Gemini report of December 2023, which introduces a family of multimodal models in Ultra, Pro and Nano sizes that work across image, audio, video and text.

  7. 2025

    Gemini Deep Think at IMO 2025

    In July 2025, an advanced version of Gemini Deep Think solves five of the six International Mathematical Olympiad problems perfectly for 35 points, a gold-medal score; Ghiasi is among the leads and key contributors of the IMO 2025 team. They also coauthor IMO-Bench, olympiad-level reasoning benchmarks published at EMNLP 2025 that played a role in that result.

  8. As of October 2026

    Aletheia at Google DeepMind

    In February 2026, they coauthor "Towards Autonomous Mathematics Research", which introduces Aletheia, a math research agent built on Gemini Deep Think that generates, verifies and revises solutions in natural language. Ghiasi is a researcher at Google DeepMind, where the Aletheia work is conducted.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

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 Golnaz Ghiasi yet. Send us a free-licensed photo.

  1. Golnaz Ghiasi - The Mathematics Genealogy Project mathgenealogy.org. Checked 19 October 2026.
  2. [1506.08347] Occlusion Coherence: Detecting and Localizing Occluded Faces arxiv.org. Checked 19 October 2026.
  3. [1605.02264] Laplacian Pyramid Reconstruction and Refinement for Semantic Segmentation arxiv.org. Checked 19 October 2026.
  4. Golnaz Ghiasi - Google Research research.google. Checked 19 October 2026.
  5. [1904.07392] NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection arxiv.org. Checked 19 October 2026.
  6. Simple Copy-Paste Is a Strong Data Augmentation Method for Instance Segmentation - CVPR 2021 Open Access Repository openaccess.thecvf.com. Checked 19 October 2026.
  7. [2112.12143] Scaling Open-Vocabulary Image Segmentation with Image-Level Labels arxiv.org. Checked 19 October 2026.
  8. [2312.11805] Gemini: A Family of Highly Capable Multimodal Models arxiv.org. Checked 19 October 2026.
  9. Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad deepmind.google. Checked 19 October 2026.
  10. Golnaz Ghiasi - ACL Anthology aclanthology.org. Checked 19 October 2026.
  11. Gemini Deep Think: Redefining the Future of Scientific Research deepmind.google. Checked 19 October 2026.
  12. Towards Autonomous Mathematics Research arxiv.org. Checked 19 October 2026.