Alokendu Mazumder

Security Science Laboratory

Fujitsu Research of India

alokendum@iisc.ac.in

I am a Senior Researcher at the Security Science (SS) Laboratory of Fujitsu Research of India. Previously, I was a Prime Minister's Research Fellow (PMRF) PhD at the Indian Institute of Science, Bengaluru, advised by Prof. Punit Rathore. My doctoral research lies at the intersection of stochastic optimization theory and deep representation learning.

I have had the privilege of conducting research at MIT's Senseable City Lab, Dolby Laboratories (Advanced Technology Group), and IBM Research — working on problems ranging from diffusion bridge models for image restoration to context-length bias in time-series forecasting.

My broader interests span optimization theory, deep learning, computer vision, and stochastic modelling, with a focus on principled, theoretically grounded approaches to modern learning problems.

Favourite Quote

“क्षमा शोभती उस भुजंग को जिसके पास गरल हो, उसको क्या जो दंतहीन विषरहित विनीत सरल हो।” रामधारी सिंह दिनकर

news

Sep 2026	LISA: Latent Inference on the Sphere for A-posteriori sampling published in Transactions on Machine Learning Research.
Jun 2026	Global Linear Convergence of Inexact TD under Generalized Smoothness accepted at Transactions on Machine Learning Research.
May 2026	On Hamming-Lipschitz Type Stability of Subdominant (Min-Max) Ultrametric: Theory and Simple Proofs accepted at Transactions on Machine Learning Research.
Apr 2026	On Convergence of ADAM with Data Dependent Stepsize accepted at IEEE Transactions on Artificial Intelligence (CiteScore 6.4).
Mar 2026	PosSpec-Net placed 4th at the ICASSP 2026 Radar Acoustic Speech Enhancement Grand Challenge.
Dec 2025	Fractional Gradient Descent with Matrix Stepsizes published in IEEE Transactions on Neural Networks and Learning Systems (Impact Factor 8.9).
May 2025	Joined MIT Senseable City Lab as a Visiting Researcher through the MIT-MISTI grant.
Mar 2025	DIME: Deterministic Information Maximizing Autoencoders accepted at the Workshop on Deep Generative Models @ ICLR 2025.
Dec 2024	Joined Dolby Laboratories (Advanced Technology Group) as a Research Intern, working on diffusion bridge models for image deblurring and dehazing.
Apr 2024	Joined IBM Research, Bengaluru as an AI Research Intern, working on context-length bias in time-series forecasting models.

selected publications

TMLR
LISA: Latent Inference on the Sphere for A-posteriori sampling

Alokendu Mazumder, Chirag Garg, Tirthajit Baruah, and 1 more author

Transactions on Machine Learning Research, Sep 2026

Abs Bib HTML PDF

LISA conditions the latent geometry of deterministic autoencoders for ex-post generation by projecting codes onto a hypersphere and encouraging directional coverage with a repulsive kernel regularizer. Coupled with projected multivariate Gaussian or Gaussian-mixture samplers, it improves generation across image benchmarks without stochastic encoders, adversarial training, or learned priors. Grammar-LISA extends the approach to structured arithmetic expressions.
@article{mazumder2026lisa, title = {{LISA}: Latent Inference on the Sphere for A-posteriori sampling}, author = {Mazumder, Alokendu and Garg, Chirag and Baruah, Tirthajit and Rathore, Punit}, year = {2026}, month = sep, journal = {Transactions on Machine Learning Research}, url = {https://openreview.net/forum?id=f0agX08y69}, }

TMLR

Global Linear Convergence of Inexact TD under Generalized Smoothness

Alokendu Mazumder, Ila Ananta, and Punit Rathore

Transactions on Machine Learning Research, 2026

Bib PDF

@article{mazumder2026td,
  title = {Global Linear Convergence of Inexact TD under Generalized Smoothness},
  author = {Mazumder, Alokendu and Ananta, Ila and Rathore, Punit},
  year = {2026},
  journal = {Transactions on Machine Learning Research},
  url = {https://openreview.net/forum?id=Xvyehi4izc},
}

TMLR

On Hamming-Lipschitz Type Stability of Subdominant (Min-Max) Ultrametric: Theory and Simple Proofs

Alokendu Mazumder, Arnab Roy, and Punit Rathore

Transactions on Machine Learning Research, 2026

Bib PDF

@article{mazumder2026ultrametric,
  title = {On Hamming-Lipschitz Type Stability of Subdominant (Min-Max) Ultrametric: Theory and Simple Proofs},
  author = {Mazumder, Alokendu and Roy, Arnab and Rathore, Punit},
  year = {2026},
  journal = {Transactions on Machine Learning Research},
  url = {https://openreview.net/forum?id=R4ASOCp3uM},
}

IEEE TAI
On Convergence of ADAM with Data Dependent Stepsize

Alokendu Mazumder, Rishabh Sabharwal, Bhartendu Kumar^*, and 4 more authors

IEEE Transactions on Artificial Intelligence, 2026

Abs Bib PDF Code

We propose a data and dynamics driven constant stepsize for Adam that removes the need for costly tuning and avoids the pitfalls of decaying schedulers. Our estimate matches the best stepsizes found by exhaustive search, performs on par with popular schedulers, and scales to ImageNet and LLM fine-tuning. We further provide theory showing that Adam with our stepsize converges to critical points for smooth non-convex objectives.
@article{mazumder2026adam, title = {On Convergence of ADAM with Data Dependent Stepsize}, author = {Mazumder, Alokendu and Sabharwal, Rishabh and Kumar, Bhartendu and Tayal, Manan and Garg, Chirag and Roy, Arnab and Rathore, Punit}, year = {2026}, journal = {IEEE Transactions on Artificial Intelligence}, url = {https://ieeexplore.ieee.org/document/11361385}, }
IEEE TNNLS
Fractional Gradient Descent with Matrix Stepsizes for Non-Convex Optimization

Alokendu Mazumder, Keshav Vyas, and Punit Rathore

IEEE Transactions on Neural Networks and Learning Systems, 2025

Abs Bib PDF Code

We establish convergence guarantees for fractional gradient descent on matrix-smooth non-convex functions and show how matrix smoothness accelerates convergence. We introduce Compressed Fractional Gradient Descent (CFGD) with matrix-valued stepsizes, proving faster convergence than scalar stepsizes in both single-node and distributed settings. This work studies fractional gradient descent in federated/distributed optimisation.
@article{mazumder2025fractional, title = {Fractional Gradient Descent with Matrix Stepsizes for Non-Convex Optimization}, author = {Mazumder, Alokendu and Vyas, Keshav and Rathore, Punit}, year = {2025}, journal = {IEEE Transactions on Neural Networks and Learning Systems}, url = {https://ieeexplore.ieee.org/document/11298544}, }

WACV

Learning Low-Rank Latent Spaces Using Simple Deterministic Autoencoders: Theoretical & Empirical Insights

Alokendu Mazumder, Tirthajit Baruah^*, Bhartendu Kumar^*, and 3 more authors

In IEEE/CVF Winter Conference on Applications of Computer Vision, 2024

Abs Bib PDF Code

@inproceedings{mazumder2024lorae,
  title = {Learning Low-Rank Latent Spaces Using Simple Deterministic Autoencoders: Theoretical \& Empirical Insights},
  author = {Mazumder, Alokendu and Baruah, Tirthajit and Kumar, Bhartendu and Sharma, Rishab and Pattanaik, Vishwajeet and Rathore, Punit},
  year = {2024},
  booktitle = {IEEE/CVF Winter Conference on Applications of Computer Vision},
  url = {https://openaccess.thecvf.com/content/WACV2024/html/Mazumder_Learning_Low-Rank_Latent_Spaces_With_Simple_Deterministic_Autoencoder_Theoretical_and_WACV_2024_paper.html},
}