Optimization & Learning Algorithms
Adaptive gradient methods, preconditioning, accelerated methods, convergence analysis, and control-theoretic perspectives on learning dynamics.
Scientist · TCS Research
Kushal
Chakrabarti
I work at the intersection of optimization, distributed learning, trustworthy AI, and control, with an emphasis on algorithms that are efficient, reliable, and supported by rigorous analysis.
About
Algorithms for intelligent systems that need to work under real constraints.
I am a Scientist at TCS Research. My research spans optimization and machine learning, distributed and federated systems, privacy and model protection, machine unlearning, and nonlinear estimation. A recurring theme is to combine ideas from optimization and control with learning systems to obtain algorithms that are not only effective in experiments, but also interpretable through convergence, robustness, communication, or protection guarantees.
Before joining TCS Research, I completed my Ph.D. in Electrical Engineering at the University of Maryland, College Park, where I worked on control-theoretic approaches to optimization and estimation. I previously earned an M.Tech. in Control and Automation from IIT Delhi and a B.E. (Hons.) in Electronics and Tele-Communication Engineering from Jadavpur University.
Research
Current research themes
The themes below reflect both my earlier work in optimization and control and my more recent work on distributed and trustworthy learning.
Distributed & Federated Learning
Communication-efficient distributed optimization, quantized algorithms, robust federated learning, and learning under network and adversarial constraints.
Trustworthy AI & Model Protection
Selective forgetting, machine unlearning, model protection against eavesdropping, and mechanisms for reliable learning in privacy-sensitive environments.
Estimation & Control
Nonlinear observers, moving-horizon estimation, consensus, networked systems, and optimization-inspired state-estimation methods.
Selected publications
Recent and representative work
2026
TMLRPublished · July 2026
On-the-go Forgetting without Explicit Unlearning via ERASE
Inference-time selective forgetting without retraining or changing the backbone model weights.
2026
AISTATS 2026Poster
Securing Model Weights Against Eavesdropping Adversaries in Federated Learning Using Quantization
2026
2026
2025
2024
AutomaticaVol. 160
A Control Theoretic Framework for Adaptive Gradient Optimizers
Recent news
Selected updates
Selected updates in publications, professional service, and recognition.
ICML Silver Reviewer Award
Recognized with an ICML 2026 Silver Reviewer Award for reviewing service.
Associate Editor · ICC 2027
Serving as an Associate Editor for the 12th Indian Control Conference, to be held at IIT Kharagpur in January 2027.
ERASE published in TMLR
“On-the-go Forgetting without Explicit Unlearning via ERASE,” with Mayank Baranwal, appears in Transactions on Machine Learning Research.
AISTATS 2026
“Securing Model Weights Against Eavesdropping Adversaries in Federated Learning Using Quantization,” with Dipankar Maity, was presented as a poster at AISTATS 2026.
Recent research across ML and control
Additional work appeared at or was accepted to IJCNN 2026, UAI 2026, CDC 2026, and in the International Journal of Robust and Nonlinear Control.
ICC tutorial on security and privacy
Co-organized the tutorial “Security and Privacy for Networked Cyber-Physical Systems” at the 11th Indian Control Conference, IISc Bengaluru, and presented on protecting distributed learning systems.
Experience & education
Research, teaching, and training
Experience
2023 — present
TCS Research
Scientist · Data & Decision Sciences
Research in optimization, distributed learning, trustworthy AI, model protection, and machine unlearning.
2023 — 2024
SP Jain School of Global Management
Adjunct Faculty · Data Science
Taught computer programming, linear algebra, and discrete mathematics.
2018 — 2022
University of Maryland
Graduate Research Assistant · Mechanical Engineering
2016 — 2017
Siksha ‘O’ Anusandhan
Assistant Professor · Electronics & Communication Engineering
Education
2022
University of Maryland, College Park
Ph.D. · Electrical Engineering
2016
Indian Institute of Technology Delhi
M.Tech. · Control & Automation
2014
Jadavpur University
B.E. (Hons.) · Electronics & Tele-Communication Engineering
Awards
ICML 2026 Silver Reviewer Award
International Conference on Machine Learning · 2026
International Conference on Machine Learning · 2026
Clark School of Engineering Distinguished Graduate Fellow
University of Maryland · 2017
University of Maryland · 2017
Talks, service & professional activities
Selected academic engagement
Associate Editor · Indian Control Conference
Associate Editor for the 12th Indian Control Conference (ICC 2027), IIT Kharagpur. Conference ↗
Security and Privacy for Networked Cyber-Physical Systems
Co-organizer and speaker, tutorial at the 11th Indian Control Conference, IISc Bengaluru.
Invited talk · IIT Bombay
Chemical Engineering, August 2024.
Accelerated Methods in Optimization: A Stability Theory Perspective
Co-organizer, tutorial at the Ninth Indian Control Conference; also co-chaired the Robotics session.
Systems & Control · IIT Bombay
Invited SysCon talk, April 2023.
Recent Trends in Modeling and Control of Dynamical Systems
Key speaker, short-term course at NIT Rourkela.
Contact
Research conversations and collaborations are welcome.
For research correspondence, either email address below is available. Google Scholar and LinkedIn provide current publication and professional updates.