PeopleofSuperintelligence

Researcher

Mohammadreza Salehi

Also known as Reza Salehi

Mohammadreza Salehi is a computer science PhD student at the University of Washington who works on open vision-language models, including Molmo and Molmo2. After a B.Sc. in computer engineering at Sharif University of Technology, he coauthored MERLOT Reserve in 2022, a model pretrained on 20 million YouTube videos, and introduced the ActionAtlas video question-answering benchmark in 2024. The Molmo and PixMo paper he coauthored received a best paper honorable mention at CVPR 2025, and in December 2025 he led long-context post-training for the video model Molmo2, which became a best paper award finalist at CVPR 2026. He works under Hannaneh Hajishirzi and Ali Farhadi and collaborates closely with Ranjay Krishna, and he started a research internship at Microsoft Research on active vision in June 2026.

Works
32
Citations
620
h-index
11
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2020

    Learning to edit paraphrases

    In 2020, he coauthors the ACL paper "Paraphrase Generation by Learning How to Edit from Samples", a retrieval-based method that finds a paraphrase pair similar to the input sentence and edits it into a new paraphrase. He earns a B.Sc. in computer engineering at Sharif University of Technology before moving to the University of Washington for his PhD.

  2. 2022

    MERLOT Reserve and ATTEMPT

    In 2022, he is a coauthor of MERLOT Reserve, a model that learns from audio, subtitles and video frames and is pretrained on 20 million YouTube videos. He also coauthors ATTEMPT in 2022, a parameter-efficient method that tunes a language model through a mixture of soft prompts.

  3. 2023

    Apple internship and SHARCS

    In January 2023, he starts a machine learning research internship with the AIML team at Apple, hosted by Sachin Mehta. In 2023, he is first author of SHARCS, a method that trains a router to direct different samples to sub-networks of varying widths in a transformer network.

  4. 2024

    Molmo and ActionAtlas

    In September 2024, he is among the authors of Molmo and PixMo, which presents Molmo, a new family of vision-language models that are state-of-the-art in their class of openness, together with the new PixMo datasets. He is also first author of ActionAtlas, a video question-answering benchmark of 934 videos across 56 sports that is accepted to NeurIPS 2024.

  5. 2025

    Student researcher at Ai2

    In March 2025, he starts as a part-time student researcher with the PRIOR team at Ai2. In 2025, the Molmo and PixMo paper appears in the CVPR 2025 proceedings and receives a best paper honorable mention.

  6. December 2025

    Molmo2 long-context post-training

    Molmo2, an open family of video-language models, comes out in December 2025. He leads long-context post-training for the model with context parallelism, and the 4B variant outperforms Qwen3-VL 8B on long video benchmarks. The Molmo2 paper presents vision-language models with point-driven grounding in single image, multi-image and video tasks.

  7. 2026

    VidNexus and a CVPR finalist

    In February 2026, he releases VidNexus, a video sourcing engine behind nearly a petabyte of video data that powers Molmo2. In 2026, Molmo2 is selected as a best paper award finalist at CVPR 2026.

  8. As of October 2026

    PhD student at Washington

    He is a sixth-year PhD student in computer science at the University of Washington, where he works under Hannaneh Hajishirzi and Ali Farhadi and collaborates closely with Ranjay Krishna. In June 2026, he starts a research internship with the AI Frontiers team at Microsoft Research, working on active vision.

Career

  1. Since 2026Microsoft ResearchResearch Intern
  2. 2025–2026Allen Institute for AI (Ai2)Student Researcher
  3. From 2023AppleMachine Learning Research Intern
  4. 2022–2025University of Amsterdam

Education

  1. 2020Sharif University of TechnologyB.Sc.
  2. University of WashingtonPhD

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. MERLOT RESERVE: Neural Script Knowledge through Vision and Language and Sound2022 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 160 citations
  2. A Unified Survey on Anomaly, Novelty, Open-Set, and Out-of-Distribution Detection: Solutions and Future Challenges2021 · 118 citations
  3. Notice of Removal: Multiresolution Knowledge Distillation for Anomaly Detection2021 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 88 citations
  4. ARAE: Adversarially robust training of autoencoders improves novelty detection2021 · Neural Networks · 60 citations
  5. Paraphrase Generation by Learning How to Edit from Samples2020 · Annual Meeting of the Association for Computational Linguistics (ACL) · 44 citations

From OpenAlex, as of 21 October 2026.

  1. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Microsoft. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  4. Google has the highest combined h-index among AI organizationsIncludes Microsoft. 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.

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 Mohammadreza Salehi yet. Send us a free-licensed photo.

  1. Reza Salehi homes.cs.washington.edu. Checked 20 October 2026.
  2. Paraphrase Generation by Learning How to Edit from Samples - ACL Anthology aclanthology.org. Checked 20 October 2026.
  3. MERLOT Reserve: Neural Script Knowledge through Vision and Language and Sound arxiv.org. Checked 20 October 2026.
  4. ATTEMPT: Parameter-Efficient Multi-task Tuning via Attentional Mixtures of Soft Prompts arxiv.org. Checked 20 October 2026.
  5. SHARCS: Efficient Transformers through Routing with Dynamic Width Sub-networks arxiv.org. Checked 20 October 2026.
  6. Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models arxiv.org. Checked 20 October 2026.
  7. ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition arxiv.org. Checked 20 October 2026.
  8. CVPR 2025 Open Access Repository openaccess.thecvf.com. Checked 20 October 2026.
  9. Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding arxiv.org. Checked 20 October 2026.