About me

Founder & CEO

Visiting Research Scientist  

Chief Scientific Officer at Ascent Robotics, Inc. *
LeapMind, Inc. Japan *
Bernstein Center for Computational Neuroscience *                      

                                                                                                                                            * previous affiliations 
E-mail: sakya [dot] lastname [at] gmail.com

News Updates

Teng, D & Dasgupta, S. (2019). Lifelong learning via Online Leverage Score Sampling. 1st Adaptive & Multitask Learning Workshop, ICML 2019. (rated in Top 50% selected as contributed talk).

Yu, P., Lee, J. S., Kulyatin, I., Shi, Z., & Dasgupta, S*. (2019). Model-based Deep Reinforcement Learning for Dynamic Portfolio Optimization. arXiv preprint arXiv:1901.08740. Real-world Sequential Decision Making Workshop ICML 2019   *correspondence

Joint work with IBM Research AI, ""Internal Model from Observations for Reward Shaping" will be presented at AAAI 2019 workshop on Reinforcement Learning in Games.

Paper on "Continuous Time-series Forecasting with Deep and Shallow Stochastic Processes" presented at NeuraIPS 2018 workshop on Continual Learning.

Paper on "Internal Model from Observations for Reward Shaping" presented at ICML 2018 Workshop on Adaptive Learning Agents (short talk).


My  paper on "Transfer learning from synthetic to real images using variational autoencoders for robotic applications" is now available on Arxiv https://arxiv.org/abs/1709.06762
Robot experiment video link: https://goo.gl/3TZP38

Our paper on "Dynamic Boltzmann Machines for Second Order Moments and Generalized Gaussian Distributions" will be presented in the Time-series workshop at NIPS 2017

Our paper on "Automated flow for compressing convolution neural networks for efficient edge-computation with FPGA" will be presented in the MLPCD workshop at NIPS 2017

Associate editor - "deep learning for robotics & automation" for top robotics conference ICRA 2018

Our paper on "Text To Image Generative Model Using Constrained Embedding Space Mapping" appears in MLSP 2017

Our paper on "
Conditional generation of multi-modal data using constrained embedding space mapping" presented at the Implicit Models workshop, ICML 2017

Delivered a half-day tutorial on "Energy-based machine learning" at IJCAI 2017 (jointly with Takayuki
Osogami ) Link: https://researcher.watson.ibm.com/researcher/view_group.php?id=7834

Dasgupta S. and Osogami T. "Nonlinear Dynamic Boltzmann Machines for Time-series Prediction" accepted at AAAI 2017 (acceptance rate of 24.6%  >2500 submissions)

Dasgupta et al. "
Regularized Dynamic Boltzmann Machine with Delay Pruning for Unsupervised Learning of Temporal Sequences", ICPR 2016 (oral presentation <14% acceptance)

Abstract selected for Talk at COSYNE 2016 main meeting.
<5% acceptance (talks) (http://www.cosyne.org)

Research Theme

My diverse research interests include nonlinear dynamical systems, recurrent neural networks, reinforcement learning, time-series modeling, computational neurodynamics, neurorobotics, optimization & control. Overall, I am interested in statistical machine learning as a field and in searching for novel theoretical frameworks to build adaptive intelligent systems.

Previously I was leading an IBM Research-Tokyo Far Reaching Research project on closed-loop experience building embodied robots learning framework (CLEBER). CLEBER is not only applicable to robots, but is a framework for online sequential decision making in an application independent manner.

Research Interests: Dynamical Systems
Energy-based machine learning

Neuro-robotics / Biorobotics
Recurrent Neural Networks
Learning and Memory
Brain Plasticity
Reinforcement Learning
Information Theory
(not necessarily in this order :))

Subpages (2): Publications Resume