PeopleofSuperintelligence

Researcher

Igor Gitman

Igor Gitman is an applied scientist at NVIDIA who focuses on mathematical reasoning in large language models. He earned a machine-learning master's degree at Carnegie Mellon in December 2017 and worked on algorithms and toolkits for GPU training. His work included large-batch neural-network training, sequence-to-sequence software, stochastic optimization and speech recognition; he later coauthored OpenMathInstruct-2 and NVIDIA's 2025 AIMO-2 winning solution with the OpenMathReasoning dataset. In 2026, he coauthored a Nemotron recipe whose system reached the International Mathematical Olympiad gold-medal threshold, and he maintains NVIDIA NeMo-Skills.

Timeline

  1. August 2017

    Large-batch convolution training

    In August 2017, he coauthors a paper that uses LARS to scale AlexNet to a batch size of 8,000 and ResNet-50 to 32,000 without loss in accuracy.

  2. December 2017

    CMU master's and NVIDIA team

    He earns a machine-learning master's degree at Carnegie Mellon in December 2017. He then works on NVIDIA's deep-learning software team, developing algorithms and toolkits for GPU training.

  3. 2018

    The OpenSeq2Seq toolkit

    He coauthors OpenSeq2Seq, an open-source toolkit for training sequence-to-sequence models with distributed and mixed-precision support.

  4. October 2019

    Momentum in stochastic gradients

    At NeurIPS in October 2019, he coauthors a unified analysis of momentum methods, covering their convergence, stability regions and stationary-distribution properties.

  5. June 2023

    Confidence-based speech ensembles

    In June 2023, he coauthors research on end-to-end speech recognition that selects the output from the most-confident model in a confidence-based ensemble.

  6. October 2024

    Building OpenMathInstruct-2

    In October 2024, he coauthors OpenMathInstruct-2, a dataset of 14 million question-solution pairs for LLM mathematical reasoning.

  7. 2025

    The AIMO-2 winning solution

    He coauthors NVIDIA's winning AIMO-2 submission and OpenMathReasoning, a dataset of 540,000 unique math problems and 3.2 million long-reasoning solutions. NVIDIA reports that the team's model answers 34 of 50 Olympiad questions. In November 2025, he coauthors a NeMo-Skills inference workflow that reaches 138.5 tokens per second with FP8 and ReDrafter on two H100 GPUs.

  8. As of October 2026

    Nemotron and NVIDIA research

    In September 2026, he coauthors an open recipe that reports the Nemotron system scores 30 of 42 points at IMO 2026 and reaches the gold-medal threshold. He is an applied scientist at NVIDIA, focused on LLM mathematical reasoning and one of the main maintainers of NeMo-Skills.

  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.

Sources and credits

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

  1. [1708.03888] Large Batch Training of Convolutional Networks arxiv.org. Checked 19 October 2026.
  2. Author: - NVIDIA 技术博客 developer.nvidia.cn. Checked 19 October 2026.
  3. OpenSeq2Seq: Extensible Toolkit for Distributed and Mixed Precision Training of Sequence-to-Sequence Models - ACL Anthology aclanthology.org. Checked 19 October 2026.
  4. Understanding the Role of Momentum in Stochastic Gradient Methods - Microsoft Research microsoft.com. Checked 19 October 2026.
  5. [2306.15824] Confidence-based Ensembles of End-to-End Speech Recognition Models arxiv.org. Checked 19 October 2026.
  6. [2410.01560] OpenMathInstruct-2: Accelerating AI for Math with Massive Open-Source Instruction Data arxiv.org. Checked 19 October 2026.
  7. [2504.16891] AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset arxiv.org. Checked 19 October 2026.
  8. NVIDIA Team Scores Kaggle Win With Reasoning Model | NVIDIA Blog blogs.nvidia.com. Checked 19 October 2026.
  9. How to Achieve 4x Faster Inference for Math Problem Solving | NVIDIA Technical Blog developer.nvidia.com. Checked 19 October 2026.
  10. [2609.10712] An Open Recipe for IMO Gold: Training Nemotron for Olympiad Mathematics arxiv.org. Checked 19 October 2026.
  11. Author: Igor Gitman | NVIDIA Technical Blog developer.nvidia.com. Checked 19 October 2026.