Contact Information
| Name | Atif Quamar |
| Professional Title | MSc Student in Machine Learning at MBZUAI |
| mohammad.atif@mbzuai.ac.ae | |
| Website | https://www.atifquamar.com |
Summary
I’m interested in reasoning and training in foundation models, particularly in understanding how reasoning abilities emerge and evolve through training. My interests span pretraining and post-training dynamics, test-time reasoning, multi-agent systems, and alignment, with a broader focus on making foundation models more capable, robust, and trustworthy.
Education
-
MSc in Machine Learning
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE
Machine Learning
-
B.Tech in Computer Science and Biosciences
Indraprastha Institute of Information Technology (IIIT), Delhi, India
Computer Science and Computational Biology
Experience
-
Graduate Student Researcher
Mohamed bin Zayed University of Artificial Intelligence
Continuing my research on efficient reasoning and training of language models.
-
Visiting Researcher
Mohamed bin Zayed University of Artificial Intelligence
Developing training and inference-time methods to improve reasoning in language models, including reinforcement learning, self-consistency, and test-time inference strategies, with applications to mathematical reasoning and coding while reducing unreliable reasoning paths.
-
Research Intern
University of Virginia
Developed a drift-resilient memory framework for code-execution agents using KL-constrained adapter updates, mitigating embedding distribution shifts during online learning to reduce unsafe code generation without compromising task success rates.
-
Research Intern
University of California San Diego
Proposed a multimodal Chain-of-Thought framework that interleaves text with latent visual representations in VLMs, with a two-stage SFT + RL training setup for modality switching, achieving strong gains across 11 multimodal reasoning benchmarks.
-
Research Intern
Purdue University
Studied Bayesian sampling and inference-time alignment methods for language models, improving response quality and steering outputs toward harmlessness and positive sentiment.
-
Co-Founder
Insituate
Founded Insituate and grew it to $250K ARR, building on-premise AI agents using open-source LLMs and RAG systems for enterprise use cases; deployments included Singapore Judiciary, Indian High Courts, Mizuho Bank, and PNC Bank.
Publications
-
2026 Reliable Chain-of-Thought via Prefix Consistency
Advances in Neural Information Processing Systems (NeurIPS) 2026
-
2026 -
2026 STARS: Synchronous Token Alignment for Robust Supervision in Large Language Models
ICML 2026 - Structured Probabilistic Inference & Generative Modeling Workshop
-
2025 Logit-Entropy Adaptive Stopping Heuristic for Efficient Chain-of-Thought Reasoning
NeurIPS 2025 - Efficient Reasoning Workshop
-
2025 -
2025 Decoding Histone Modification Signatures of Non-Coding RNAs via Foundation Models
NeurIPS 2025 - FM4LS Workshop