Mohammad Azhar Khan

I am an M.Tech student in Computer Science and Engineering at IIT Hyderabad and currently working as a Research Assistant in the domain of Theoretical Machine Learning and Scalable Data Science.

I am working under Professor Dr. Rameshwar Pratap, focusing on advancing tensor sketching techniques for efficient, high-fidelity dimensionality reduction. My work involves rigorous theoretical analysis of sketching algorithms, including variance bounds, computational complexity analysis, and accuracy guarantees.

My research aims at overcoming the Curse of Dimensionality in machine learning models using advanced sketching and random projection techniques. I contribute to algorithm design, empirical benchmarking, and manuscript preparation targeting top-tier Machine Learning and Theoretical Computer Science conferences/journals.

Previously, as a Research Intern at SPIRE Lab, IISc Bangalore, I worked on language models and designed RNN-based NLP systems for grammatical error detection and pronunciation assessment.

Indian Institute of Technology, Hyderabad | 2024 - 2027

Master of Technology (M.Tech) in Computer Science and Engineering

Research Assistant | CGPA: 8.00 / 10.00

Government College Of Engineering, Keonjhar | 2020 - 2024

Bachelor of Technology (B.Tech) in Computer Science and Engineering

CGPA: 8.86 / 10.00

Sri Chaitanya Techno School, Visakhapatnam | 2020

Higher Secondary Education (12th Grade)

Grade: 83%

Efficient and Accurate Tensor Compression via Recursive Sketching

Amit Sharma, Mohammad Azhar Khan, Rameshwar Pratap

The 29th International Conference on Artificial Intelligence and Statistics (AISTATS 2026)

We propose improved tensor sketching algorithms for estimating pairwise inner products of high-order tensor data. Our method provides unbiased estimates with significantly lower variance—independent of the number of tensor modes— improving upon prior work by Rakhshan and Rabusseau (AISTATS 2020). Additionally, our recursive sketching framework achieves asymptotically improved time complexity, building upon techniques introduced by Ahle et al. (SODA 2020).

View Paper (OpenReview)

IIT Hyderabad | Research Assistant | 2024 - Present

• Advancing Tensor Sketching methods for efficient, high-fidelity dimensionality reduction of tensors and related topics.
• Specializing in the theoretical analysis of sketching algorithms, including rigorous proofs for variance bounds, computational complexity, and accuracy guarantees.
• Investigating methods to overcome the Curse of Dimensionality in ML models through advanced sketching and random projection techniques.
• Contributing to algorithm design, empirical benchmarking with code, and manuscript preparation for top-tier Machine Learning and Theoretical Computer Science conferences/journals.

SPIRE Lab, IISc Bangalore | Research Intern | Dec 2022 - April 2023

• Worked on language models and designed an NLP model based on RNN which can point out Grammatical Errors in the sentences.
• Designed a mechanism for an mobile application English Gyani which can rate human pronunciation.