About me

My research laid the foundations for Diffusion Language Models, and I now lead a team at MBZUAI - IFM advancing the field. My work is used at industrial scale by Google, NVIDIA, and ByteDance across applications from language generation to drug discovery.

Ph.D. Thesis: Foundations of Diffusion Language Models, advised by Prof. John Thickstun.
Previously: Cornell Tech (Ph.D.); Google Research; IIT Kharagpur (B.Tech).

Highlights

Select Papers

  • Subham S. Sahoo, Justin Deschenaux, Aaron Gokaslan, Guanghan Wang, Justin Chiu, Volodymyr Kuleshov. The Diffusion Duality. 42nd International Conference on Machine Learning (ICML 2025), ICLR 2025 - DeLTa Workshop (oral). [paper, code, webpage]


    Subham S. Sahoo, Marianne Arriola, Yair Schiff, Aaron Gokaslan, Edgar Marroquin, Justin T Chiu, Alexander Rush, Volodymyr Kuleshov. Simple and Effective Masked Diffusion Language Models. 38th Conference on Neural Information Processing Systems (NeurIPS 2024), ICML 2024 - AccMLBio Workshop (spotlight). [paper, code, webpage]


    Subham S. Sahoo, Aaron Gokaslan, Chris De Sa, Volodymyr Kuleshov. Diffusion Models With Learned Adaptive Noise. 38th Conference on Neural Information Processing Systems (NeurIPS 2024, spotlight), NeurIPS 2024 - Compression Workshop (spotlight). [paper, code, webpage]


    Subham S. Sahoo*, Anselm Paulus*, Marin Vlastelica, Vit Musil, Volodymyr Kuleshov, Georg Martius. Backpropagation through Combinatorial Algorithms: Identity with Projection Works. 11th International Conference on Learning Representations (ICLR 2023). [paper, code]


    Subham S. Sahoo, Christoph H. Lampert, Georg Martius. Learning Equations for Extrapolation and Control. 35th International Conference on Machine Learning (ICML 2018). [paper, code, webpage]

News

  • Mar-3-26: Eso-LMs accepted at ICLR 2026, Workshop on Multimodal Intelligence as an oral!

    Jan-26-26: The Diffusion Duality, Chapter II: Ψ-Samplers accepted at ICLR 2026!

    Oct-23-25: Invited talk at Radboud University on The Diffusion Duality.

    Oct-15-25: Invited talk at Seoul National University on Foundations of Diffusion Language Models. [slides]

    Oct-3-25: Defended my Ph.D. Thesis: Foundations of Diffusion Language Models. [slides]

    Aug-13-25: Invited talk at Cerebras on Esoteric Language Models.

    Aug-6-25: Invited talk at Meta (FAIR) on Foundations of Diffusion Language Models.

    Jun-19-25: Invited talk at Google Deepmind on Esoteric Language Models.

    May-1-25: Duo accepted at ICML 2025!

    Apr-28-25: Presenting Duo as an oral at ICLR 2025, DeLTa workshop!

    Apr-2-25: Invited talk at Databricks on Diffusion Language Models.

    Mar-24-25: Invited talk at Genesis Therapeutics on Diffusion Language Models.

    Mar-19-25: Invited for Research/Industrial Inference/PostTraining focused Round Table at Nvidia GTC-2025.

    Mar-7-25: Invited talk at Nvidia on Diffusion Language Models. [slides]

    Feb-11-25: BD3-LM and UDLM accepted at ICLR 2025! BD3-LM has been accepted as an oral!

    Dec-10-24: MDLM and MuLAN accepted at NeurIPS 2025! MuLAN was presented as a spotlight!

    Oct-11-24: Passed my Ph.D. Candidacy exam!

    Jul-27-24: Presented MDLM as a spotlight at ICML 2024, AccMLBio workshop!

Daniel lewis
quote icon

Daniel lewis

Richard was hired to create a corporate identity. We were very pleased with the work done. She has a lot of experience and is very concerned about the needs of client. Lorem ipsum dolor sit amet, ullamcous cididt consectetur adipiscing elit, seds do et eiusmod tempor incididunt ut laborels dolore magnarels alia.

Papers

  • Justin Deschenaux, Caglar Gulcehre, Subham S. Sahoo. The Diffusion Duality, Chapter II: Ψ-Samplers and Efficient Curriculum. 14th International Conference on Learning Representations (ICLR 2026). [paper, code, webpage]


    Pin-Jui Ku, He Huang, Jean-Marie Lemercier, Subham S. Sahoo, Zhehuai Chen, Ante Jukic. Discrete Diffusion for Generative Modeling of Text-Aligned Speech Tokens. IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2026). [paper]


    Subham S. Sahoo*, Zhihan Yang*, Yash Akhauri, Johnna Liu, Deepansha Singh, Zhoujun Cheng, Zhengzhong Liu, Eric Xing, John Thickstun, Arash Vahdat. Esoteric Language Models. Pre-print. [paper, code, webpage]


    Subham S. Sahoo,Jean-Marie Lemercier, Zhihan Yang, Justin Deschenaux, Jingyu Liu, John Thickstun, Ante Jukic. Scaling Beyond Masked Diffusion Language Models. Pre-print. [paper, code, webpage]

  • Subham S. Sahoo, Justin Deschenaux, Aaron Gokaslan, Guanghan Wang, Justin Chiu, Volodymyr Kuleshov. The Diffusion Duality. 42nd International Conference on Machine Learning (ICML 2025), ICLR 2025 - DeLTa Workshop (oral). [paper, code, webpage]


    Guanghan Wang, Yair Schiff, Subham S. Sahoo, Volodymyr Kuleshov. Remasking Discrete Diffusion Models with Inference-Time Scaling. 39th Conference on Neural Information Processing Systems (NeurIPS 2025), ICLR 2025 - DeLTa Workshop. [paper, code, webpage]


    Marianne Arriola, Subham S. Sahoo, Aaron Gokaslan, Zhihan Yang, Zhixuan Qi, Jiaqi Han, Justin Chiu, Volodymyr Kuleshov. Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models. 13th International Conference on Learning Representations (ICLR 2025, oral). [paper, code, webpage]


    Yair Schiff*, Subham S. Sahoo*, Hao Phung*, Guanghan Wang*, Sam Boshar, Hugo Dalla-torre, Bernardo P de Almeida, Alexander M Rush, Thomas Pierrot, Volodymyr Kuleshov. Simple Guidance Mechanisms for Discrete Diffusion Models. 13th International Conference on Learning Representations (ICLR 2025). [paper, code, webpage]

  • Subham S. Sahoo, Marianne Arriola, Yair Schiff, Aaron Gokaslan, Edgar Marroquin, Justin T Chiu, Alexander Rush, Volodymyr Kuleshov. Simple and Effective Masked Diffusion Language Models. 38th Conference on Neural Information Processing Systems (NeurIPS 2024), ICML 2024 - AccMLBio Workshop (spotlight). [paper, code, webpage]


    Subham S. Sahoo, John X. Morris, Aaron Gokaslan, Srijeeta Biswas, Vitaly Shamtikov, Volodymyr Kuleshov. Zero-Order Diffusion Guidance for Inverse Problems. Pre-print. [paper]


    Subham S. Sahoo, Aaron Gokaslan, Chris De Sa, Volodymyr Kuleshov. Diffusion Models With Learned Adaptive Noise. 38th Conference on Neural Information Processing Systems (NeurIPS 2024, spotlight), NeurIPS 2024 - Compression Workshop (spotlight). [paper, code, webpage]

  • Subham S. Sahoo, Anselm Paulus, Marin Vlastelica, Vit Musil, Volodymyr Kuleshov, Georg Martius. Backpropagation through Combinatorial Algorithms: Identity with Projection Works. 11th International Conference on Learning Representations (ICLR 2023). [paper, code]


    Phillip Si, Zeyi Chen, Subham S. Sahoo, Subham S. Sahoo, Yair Schiff, Volodymyr Kuleshov. Semi-Autoregressive Energy Flows: Towards Determinant-Free Training of Normalizing Flows. 40th International Conference on Machine Learning (ICML 2023). [paper]

  • Subham S. Sahoo, Subhashini Venugopalan, Li Li, Rishabh Singh, Patrick Riley. Scaling Symbolic Methods using Gradients for Neural Model Explanation. 9th International Conference on Learning Representations (ICLR 2021). [paper, code]


    Subham S. Sahoo, Ross Anderson, Christian Tjandraatmadja. Local Search on TPUs. Pre-print, 2021. [paper]

  • Subham S. Sahoo. Training Neual Networks using SAT solvers. Pre-print, 2018. [paper]


    Subham S. Sahoo, Christoph H. Lampert, Georg Martius. Learning Equations for Extrapolation and Control. 35th International Conference on Machine Learning (ICML 2018). [paper, code, webpage]

Panels & Talks

  • Panels

    Jul-09-26: Diffusion Language Models vs Autoregressive Language Models at ICML 2026 Workshop on Structured Probabilistic Inference & Generative Modeling.


    Mar-19-25: Research/Industrial Inference/PostTraining focused Round Table at Nvidia GTC-2025.

  • Contributed Talks

    Jun-03-26: At CVPR 2026, Tutorial : "The Principles of Diffusion Models: Real-Time Continuous & Discrete Diffusion".


    Apr-28-24: At ICLR 2026 - Workshop on Multimodal Intelligence Workshop, "Esoteric Language Models".


    Apr-28-24: At ICLR 2025 - DeLTA Workshop, "The Diffusion Duality".


    Dec-15-24: At NeurIPS 2024 - Compression Workshop, "Diffusion Models with Learned Adaptive Noise". [slides]


    Jul-27-24: At ICML 2024 - AccMLBio Workshop, "Simple and Effective Masked Diffusion Language Models". [slides].

  • Invited Talks

    Oct-23-25: At Radboud University , "The Diffusion Duality".


    Oct-15-25: At Seoul National University , "Foundations of Diffusion Language Models".


    Aug-13-25: At Cerebras , "Esoteric Language Models".


    Aug-6-25: At Meta , "Foundations of Diffusion Language Models".


    Jun-19-25: At Google Deepmind , "Esoteric Language Models".


    Apr-2-25: At Databricks , "Diffusion Language Models". [slides]


    Mar-24-25: At Genesis Therapeutics, "Simple and Effective Masked Diffusion Language Models".


    Mar-7-25: At Nvidia, "Diffusion Language Models". [slides]

Background

Experience

  1. MBZUAI - Institute of Foundation Models, San Francisco, USA.

    2026 - Present

    Sr. Research Scientist (Team Lead).
    Team: Diffusion Language Models.

  2. Cruise, San Francisco, USA.

    2023 (May - July)

    Research intern.
    Team: AV Behaviors.

  3. Max Planck Institute for Intelligent Systems, Tubingen, Germany.

    2021 (Aug - Dec)

    Visiting Researcher.
    Team: Autonomous Learning Group.

  4. Google Research, Mountain View, USA.

    2019 — 2021

    AI Resident.
    Teams: Accelerated Science, Operations Research.

Education

  1. Cornell Tech, New York, USA.

    2022 — 2025

    Ph.D. in Computer Science.
    Thesis: Foundations of Diffusion Language Models
    Committee: Prof. John Thickstun (chair), Prof. Noah Snavely, Prof. Bart Selman.

  2. Indian Institute of Technology - Kharagpur, India.

    2015 — 2019

    Bachelor's in Electrical Engineering.