Atif Quamar

About

I am currently a Master’s student in Machine Learning at MBZUAI, advised by Dr. Junpei Komiyama. My research lies at the intersection of foundation model training and mechanistic interpretability, focusing on pretraining and post-training dynamics, machine reasoning, and multi-agent systems. I am particularly interested in how reasoning and internal mechanisms emerge during training, and how understanding them can lead to more reliable, robust, and aligned models.

Previously, I worked on latent reasoning in vision-language models with Dr. Biwei Huang at UC San Diego and on inference-time alignment of language models with Dr. Berkay Celik at Purdue University. I also had the opportunity to collaborate with Dr. Ferdinando Fioretto at the University of Virginia on understanding misalignment in self-evolving coding agents during iterative self-improvement.

News

Sep 2026	Our work on Prefix-Consistency was accepted to NeurIPS 2026!
Sep 2026	Attended the ELLIS Summer School at TU Munich, Germany!
Aug 2026	Our new work on Prefix-Denoising Consistency is now up on arXiv!
Aug 2026	Started MSc in Machine Learning at MBZUAI!
Aug 2026	Our new work on On-Policy Self-Distillation is now up on arXiv!
May 2026	STARS was accepted at the SPIGM Workshop at ICML 2026!
May 2026	Our new work on Prefix-Consistency is now up on arXiv!
Feb 2026	Joined MBZUAI as a Visiting Researcher to work on reasoning in language models!
Oct 2025	Served on the Program Committee for NeurIPS 2025 Workshops!
Oct 2025	Two papers accepted at NeurIPS workshops — Efficient Reasoning and FM4LS!
Aug 2025	Served on the Program Committee for AAAI 2026!

Publications ^* indicates equal contribution.

NeurIPS 2026

Reliable Chain-of-Thought via Prefix Consistency

Naoto Iwase , Yuki Ichihara , Mohammad Atif Quamar , and Junpei Komiyama

Advances in Neural Information Processing Systems (NeurIPS) 2026

Project Paper Bib

@misc{iwase2026reliablechainofthoughtprefixconsistency,
  title = {Reliable Chain-of-Thought via Prefix Consistency},
  author = {Iwase, Naoto and Ichihara, Yuki and Quamar, Mohammad Atif and Komiyama, Junpei},
  year = {2026},
  eprint = {2605.07654},
  archiveprefix = {arXiv},
  primaryclass = {stat.ML},
  url = {https://arxiv.org/abs/2605.07654},
}

Preprint

Adaptive Blockwise Search: Inference-Time Alignment for Large Language Models

Mohammad Atif Quamar^* , Mohammad Areeb^* , Nishant Sharma , Ananth Shreekumar , Jonathan Rosenthal , Muslum Ozgur Ozmen , Mikhail Kuznetsov , and Z. Berkay Celik

arXiv preprint, 2025

Paper Code Bib

@misc{quamar2025adaptiveblockwisesearchinferencetime,
  title = {Adaptive Blockwise Search: Inference-Time Alignment for Large Language Models},
  author = {Quamar, Mohammad Atif and Areeb, Mohammad and Sharma, Nishant and Shreekumar, Ananth and Rosenthal, Jonathan and Ozmen, Muslum Ozgur and Kuznetsov, Mikhail and Celik, Z. Berkay},
  year = {2025},
  eprint = {2510.23334},
  archiveprefix = {arXiv},
  primaryclass = {cs.CL},
  url = {https://arxiv.org/abs/2510.23334},
}

NeurIPS-W 2025

Logit–Entropy Adaptive Stopping Heuristic for Efficient Chain-of-Thought Reasoning

Mohammad Atif Quamar and Mohammad Areeb

NeurIPS 2025 Workshop on Efficient Reasoning

Paper OpenReview Bib

@inproceedings{quamar2025logitentropy,
  title = {Logit{\textendash}Entropy Adaptive Stopping Heuristic for Efficient Chain-of-Thought Reasoning},
  author = {Quamar, Mohammad Atif and Areeb, Mohammad},
  booktitle = {NeurIPS 2025 Workshop on Efficient Reasoning},
  year = {2025},
  url = {https://openreview.net/forum?id=5UTXI0iNn5},
}

ICML-W 2026

STARS: Synchronous Token Alignment for Robust Supervision in Large Language Models

Mohammad Atif Quamar^* , Mohammad Areeb^* , Mikhail Kuznetsov , Muslum Ozgur Ozmen , and Z. Berkay Celik

ICML 2026 Workshop on Structured Probabilistic Inference & Generative Modeling

Paper Code Bib

@inproceedings{quamar2026stars,
  title = {{STARS}: Synchronous Token Alignment for Robust Supervision in Large Language Models},
  author = {Quamar, Mohammad Atif and Areeb, Mohammad and Kuznetsov, Mikhail and Ozmen, Muslum Ozgur and Celik, Z. Berkay},
  booktitle = {ICML 2026 Workshop on Structured Probabilistic Inference {\&} Generative Modeling},
  year = {2026},
  url = {https://openreview.net/forum?id=M3BAEZGcnu},
}

Preprint

Privileged Solutions or Context-Induced Teacher Behavior? Dissecting On-Policy Self-Distillation

Yuki Ichihara , Naoto Iwase , Mohammad Atif Quamar , and Junpei Komiyama

arXiv preprint, 2026

Paper Code Bib

@misc{ichihara2026privilegedsolutionscontextinducedteacher,
  title = {Privileged Solutions or Context-Induced Teacher Behavior? Dissecting On-Policy Self-Distillation},
  author = {Ichihara, Yuki and Iwase, Naoto and Quamar, Mohammad Atif and Komiyama, Junpei},
  year = {2026},
  eprint = {2608.09228},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2608.09228},
}

Preprint

Prefix-Denoising Consistency: Test-Time Verification for Diffusion Language Models

Yuki Ichihara , Naoto Iwase , Mohammad Atif Quamar , and Junpei Komiyama

arXiv preprint, 2026

Paper Code Bib

@misc{ichihara2026prefixdenoisingconsistencytesttimeverification,
  title = {Prefix-Denoising Consistency: Test-Time Verification for Diffusion Language Models},
  author = {Ichihara, Yuki and Iwase, Naoto and Quamar, Mohammad Atif and Komiyama, Junpei},
  year = {2026},
  eprint = {2608.25311},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2608.25311},
}

Preprint

Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings

Yifei Shao , Kun Zhou , Ziming Xu , Mohammad Atif Quamar , Shibo Hao , Zhen Wang , Zhiting Hu , and Biwei Huang

arXiv preprint, 2026

Paper Bib

@misc{shao2026learningmodalmixedchainofthoughtreasoning,
  title = {Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings},
  author = {Shao, Yifei and Zhou, Kun and Xu, Ziming and Quamar, Mohammad Atif and Hao, Shibo and Wang, Zhen and Hu, Zhiting and Huang, Biwei},
  year = {2026},
  eprint = {2602.00574},
  archiveprefix = {arXiv},
  primaryclass = {cs.AI},
  url = {https://arxiv.org/abs/2602.00574},
}

NeurIPS-W 2025

Decoding Histone Modification Signatures of Non-Coding RNAs via Foundation Models

Nishant Sharma , Mohammad Atif Quamar , and Pengtao Xie

NeurIPS 2025 2nd Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences

OpenReview Bib

@inproceedings{sharma2025decoding,
  title = {Decoding Histone Modification Signatures of Non-Coding {RNA}s via Foundation Models},
  author = {Sharma, Nishant and Quamar, Mohammad Atif and Xie, Pengtao},
  booktitle = {NeurIPS 2025 2nd Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences},
  year = {2025},
  url = {https://openreview.net/forum?id=SIVkGvPDuM},
}

About

News

Publications * indicates equal contribution.

Publications ^* indicates equal contribution.