A more complete list of publications can be found
Towards Interpretable Reinforcement Learning: Interactive
Visualizations To Increase Insight
Experimentation, design, and prototype of a system to increase insight into reinforcement learning algorithms and agent policies through interactive visualizations. Discussion of a related research investigation to increase insight into the learning dynamics of actor-critic learning algorithms by visualizing optimization landscapes.
[NeurIPS 2020 Deep Reinforcement Learning Workshop]
Design and implementation of an interactive visualization tool for
debugging and interpreting reinforcement learning algorithms.
Vizarel: A System to Help Better Understand RL Agents
( Paper )
S Deshpande, J Schneider
[(Spotlight Talk) ICML 2020 Workshop on Human
Interpretability in Machine Learning (WHI)]
Prototype of a visualization tool for
reinforcement learning researchers to interpret, introspect, and
gain intuition for the successes and failures of trained policies.
Learning Radiative Transfer Models for Climate Change
Applications in Imaging Spectroscopy
S Deshpande, BD Bue, DR Thompson, V Natraj, M Parente
[ICML 2019 Workshop. Climate Change: How Can AI Help?]
Designed neural network architectures to emulate differential
equation solvers. Primary motivation was to help better understand
climate change effects on coral reefs. Work led to 10x improvement
over previous known solution.
Neural network radiative transfer for imaging spectroscopy
BD Bue, DR Thompson, S Deshpande, M Eastwood, RO Green, V Natraj,
T Mullen, M Parente
[(Journal) Atmospheric Measurement Techniques]
Demonstrated that nonparametric function approximation with neural
networks can replicate radiative transfer calculations, generate
accurate radiance spectra at multiple wavelengths over a diverse
range of surface and atmosphere state parameters, and can act as
surrogate forward models for atmospheric correction procedures.
Evaluated the approach in atmospheric correction of data from the
PRISM airborne imaging spectrometer.
S Deshpande*, R Desai*, A Sakhadeo*, S Shaikh*, Y Wadadekar, H
Intema, B Ratnakumar, L George
[(Lightning Talk) ADASS XXVIII] [(Journal) Publications of the Astronomical Society of the
Built a compute cluster and data science pipeline to synthesize
images from a radio telescope. Effort reduced synthesis time from 6+
weeks to 10-12 hours. Project is currently generating one of the
world's largest catalogs of sub-GHz frequency radio astronomy
Spinet: Analyzing Statistical Properties Of Data In Large Scale
S Deshpande, A Nambi, V Padmanabhan
Explored the use of streaming algorithms to monitor data center
network traffic. Potential future tech transfer to Microsoft Azure.
India Digital Heritage Project: Designing An Immersive VR
Worked on optimizing a VR rendering pipeline for mobile devices.
Broader effort was to preserve heritage monuments through 3D laser
scans. Got to present to the Secretary of Science and Technology
(Govt. of India), and Secretary for Ministry of Culture :)