Avishek Ghosh
Assistant Professor, Indian Institute of Technology, Bombay (IITB)
Contact
Office: Room 110, SysCon
Email: avishek_ghosh@iitb.ac.in, avishek_ghosh@berkeley.edu
Research Interests: I work in Theoretical Machine Learning including
Optimization - Federated Learning (FL), Incentivised FL, Mechanism design for FL
Online Learning and Multi-armed Bandits - Reinforcement Learning, anomaly detection in online learning
Learning Theory - Mixtures, Model Selection
About me: I am an Assistant Professor in Systems and Control Engineering (SysCon) and Centre for Machine Intelligence and Data Science (C-MInDS) (newly built ML Dept.) at the Indian Institute of Technology, Bombay. I work in Theoretical Machine Learning with a focus on (a) Large Scale Optimization in the Federated Learning framework and (b) Sequential learning in Reinforcement Learning and Multi Armed Bandits framework.
Previously, I was a Postdoctoral fellow at HDSI (Data-Science Institute), University of California, San-Diego, where I worked with Prof. Arya Mazumdar and Prof. Tara Javidi. Prior to that, I did my PhD from the EECS Dept. of University of California, Berkeley, with Prof. Kannan Ramchandran (EECS) and Prof. Aditya Guntuboyina (Statistics). Moreover, I spent the summer of 2020 working at Amazon Research New York, where I had the opportunity to work with Dean Foster and Prof. Sasha Rakhlin. I did my masters in the ECE Dept. of IISc Bangalore (working with Prof. Anurag Kumar and Prof. Aditya Gopalan) and Bachelors from the ETCE Dept. of Jadavpur University.
Students interested in working with me: If you want to work in Theoretical Machine Learning and Statistical Optimization, I encourage you to get in touch with me. You can work with me through either Systems and Control Engineering (SysCon) or Centre for Machine Intelligence and Data Science (C-MInDS). Feel free to send an email for further details. Also, have a look at my research page for more information.
Currently looking for students with an interest in:
Multi-armed Bandits (MAB)
Federated Learning
Non-stationary and robust online learning
Updates:
May 24 -- Paper on Agnostic Learning of Mixed Linear Regressions with EM and AM Algorithms accepted at International Conference on Machine Learning (ICML), 2024, Vienna, Austria
May 24 -- Paper on PairNet: Training with Observed Pairs to Estimate Individual Treatment Effect accepted at International Conference on Machine Learning (ICML), 2024, Vienna, Austria
April 24 -- Paper on Explore-then-Commit Algorithms for Decentralized Two-sided Matching Markets accepted at IEEE International Symposium on Information Theory (ISIT) 2024
April 24 -- Paper on saddle point avoidance using cubic regularized Newton method accepted at IEEE International Symposium on Information Theory (ISIT) 2024
April 24 -- Paper on False data injection in cyber-physical system accepted at IEEE International Symposium on Information Theory (ISIT) 2024
Feb 24 -- [Invited] Talk at the Reinforcement Learning: Recent Trends and Future Challenges Workshop at IISc Bangalore
Feb 24 -- [Invited] Talk at TIFR Mumbai on Non Stationary Matching Markets
Feb 24 -- Paper on Improved Federated Clustering accepted at Transactions in Machine Learning Research (TMLR)
Dec 23 -- Paper on Dynamic matching markets accepted at IEEE Transactions on Information Theory
Dec 23 -- Organized Indian Symposium on Machine Learning (IndoML) at IIT Bombay.
Dec 23 -- Organized workshop on Large Scale Learning Learning and Control at IIT Bombay. Please check this link
Sept 23 -- Proposal on Incentivizing Contribution in Federated Learning got accepted at Amazon IITB AI-ML Initiative program (Co-PI: Swaprava Nath, CSE, IITB)
Jul 23 -- Paper on Model selection for Generic contextual bandits accepted at IEEE Transactions on Information Theory
Jul 23 -- Paper on Understanding and Control of Zener Pinning via Phase Field and Ensemble Learning accepted at Computational Materials Science
Jun 23 -- 2 papers on Economic Markets and (Multi-armed) Bandits (on dynamic markets and two sided competitive learning) accepted at ICML 2023 (Many Facets of Preference-Based Learning Workshop)
Jun 23 -- 2 papers on Improved Clustering in Federated Learning and Saddle point avoidance accepted at ICML 2023 (Federated Learning and Analytics Workshop)
April 23 -- Paper on Optimal Compression of Unit Norm Vectors in the High Distortion Regime accepted in ISIT 2023
March 23 -- Talks on Federated Learning and Non Convex Optimization at Electrical Engg. and Computer Science respectively, IIT Bombay
Jan 23 -- Paper on Spectral Lower Bounds for Linear Bandits accepted in AISTATS 2023
Dec 22 -- Joined IIT Bombay as an Assistant Professor
Nov 22 -- New paper on Improved Clustered Federated Learning arxived: https://arxiv.org/abs/2210.11538
Sept 22 -- Presented 2 papers (long-talk) at ECML-PKDD 2022
July 22 -- Presented 2 Spotlight talks at ICML 2022, Baltimore
July 22 -- Paper on Clustered Federated Learning (long version) accepted in IEEE Transactions on Information Theory
June 22 -- Paper on Model Selection in Reinforcement Learning with General Function Approximations accepted at ECML-PKDD 2022
June 22 -- Paper on Multi-Agent Heterogeneous Stochastic Linear Bandits (Clustering and Personalization) accepted at ECML-PKDD 2022
May 22 -- Paper on Breaking the $\sqrt{T}$ Barrier: Instance Independent Logarithmic Regret for Contextual Bandits accepted at ICML 2022
May 22 -- Paper on Learning Mixture of Linear Regressions in the Non-Realizable Setting accepted at ICML 2022
Jan 22 -- Talk on Federated Learning; Communication Efficiency and Robustness; The Institute for Learning-enabled Optimization at Scale -- The Institute for Learning-enabled Optimization at Scale (TILOS) virtual meet.
Nov -21: Paper on Max affine regression for Gaussian design got accepted at IEEE Transactions on Information Theory
Aug -21: Paper on Communication efficient Byzantine Resilience in Federated Learning with Error Feedback got accepted at IEEE Journal on Selected Areas in Information Theory
May -21: Paper on Distributed Newton in Federated Learning with local averaging got accepted at UAI 2021
May -21: Joined Halıcıoğlu Data Science Institute (HDSI), UC San Diego as a Data Science Postdoctoral Fellow
April - 21: 2 posters at the NSF-TRIPODS workshop on Communication Efficient and Distributed Optimization
Jan -21: Paper on model selection for stochastic contextual bandits got accepted at AISTATS 2021
Sep - 20: 2 Papers on Clustered Federated Learning and Comm. efficient and Byzantine robust distributed Newton accepted at NeurIPS 2020.
Selected (10) Publications (see full list here):
Agnostic Learning of Mixed Linear Regressions with EM and AM Algorithms; Avishek Ghosh and Arya Mazumdar; ICML, 2024
Decentralized Competing Bandits In Non-Stationary Matching Markets; Avishek Ghosh, Abishek Sankararaman, Kannan Ramchandran, Tara Javidi and Arya Mazumdar-- IEEE Transactions on Information Theory, 2024.
Breaking the $\sqrt{T}$ Barrier: Instance-Independent Logarithmic Regret in Stochastic Contextual Linear Bandits; Avishek Ghosh and Abishek Sankararaman -- ICML, 2022
On Learning Mixture of Linear Regressions in the Non-Realizable Setting; Avishek Ghosh, Arya Mazumdar, Soumyabrata Pal and Rajat Sen -- ICML, 2022
An Efficient Framework for Clustered Federated Learning--Avishek Ghosh, J. Chung, Dong Yin, Kannan Ramchandran--NeurIPS 2020 (long version in IEEE Transactions on Information Theory, 2022)
Max-Affine Regression: Parameter Estimation for Gaussian Designs; Avishek Ghosh, Ashwin Pananjady, Aditya Guntuboyina, Kannan Ramchandran--IEEE Transactions on Information Theory, 2022
Problem-Complexity Adaptive Model Selection for Stochastic Linear Bandits -- Avishek Ghosh, Abishek Sankararaman, Kannan Ramchandran--AISTATS 2021
Alternating Minimization Converges Super-linearly for Mixed Linear Regression; Avishek Ghosh, Kannan Ramchandran--AISTATS 2020
Communication-Efficient and Byzantine-Robust Distributed Learning with Error Feedback; Avishek Ghosh, Raj Kumar Maity, Swanand Kadhe, Arya Mazumdar and Kannan Ramchandran, IEEE Journal on Selected Areas in Information Theory, 2021 (a variation in NeurIPS 2020).
Misspecified Linear Bandits ; Avishek Ghosh, Sayak R. Chowdhury, Aditya Gopalan--AAAI 2017
Current Collaborators
Arya Mazumdar , Professor, University of California, San Diego
Kannan Ramchandran, Professor, University of California, Berkeley
Sunita Sarawagi, Professor, CSE, IIT Bombay
Nikhil Karamchandani, Professor, EE, IIT Bombay
Swaprava Nath, Professor, CSE, IIT Bombay
Vivek Borkar, Professor, EE, IIT Bombay
Abishek Sankararaman, Research Scientist, Amazon, USA
Dong Yin, Research Scientist, Deepmind (Google), USA
Aditya Gopalan, Professor, Indian Institute of Science, Bangalore
Sayak Ray Chowdhury, Postdoc, Microsoft Research, India
Reviewer
Neural Information Processing Systems (NeurIPS) -- 2020, 2021, 2022
International Conference on Machine Learning (ICML) -- 2020, 2021 (recipient of top 33% Reviewer Certificate), 2022, 2023
Journal of Machine Learning Research (JMLR) (Journal)
International Conference on Learning Representations (ICLR), 2021, 2022
International Conference on Artificial Intelligence and Statistics (AISTATS), 2021, 2022, 2023, 2024
IEEE Transactions on Information Theory (Journal)
IEEE Transactions on Neural Networks and Learning Systems (Journal)
International Symposium on Information theory (ISIT) -- 2020, 2021,2023
Machine Learning, Springer (Journal)
IEEE Transactions on Signal Processing (Journal)
IEEE/ACM Transactions on Networking (Journal)
Signal Processing (Journal)
Computational Statistics and Data Analysis (Journal)
IEEE Conference on Decision and Control (CDC), 2023