Hi, I'm Aditya, a Machine Learning Researcher at San Diego. Trying to make the world a better a place for all!

Generating Dense Labels for Earth Observation using GAN's

In this project we, at Radiant Earth Foundation, worked on producing semantic labels for medium resolution earth observation obtained from Sentinel-2 satellites. Having per pixel annotation of earth observation is of great advantage in crafting ecological and economic policies, but these are extremely labor intensive to get. The wide diversity of land type, seasons across the world demands availability of labeled data everywhere; making it even harder to use machine learning based approaches. In this project we developed conditional GAN's to perform the task of annotation and achieved the superior performance over widely used deep learning method involving CNN's when we moved to unseen test geographies.

Our work was published in NeurIPS 2020 AI for Earth Workshop: https://arxiv.org/abs/2012.03093

Making Energy Supply Sustainable

In this project we, at Renewable Energy Lab UC San Diego, are trying to make very accurate prediction of electric ramps across California. Electric ramp is a measure of deficit energy demand that is not met by renewables like solar and/or wind sources. Using these prediction enables better integration of renewables like Solar and Wind energy supplies into the current electric grids. Currently our focus is on the data made available by CAISO and EIA for the California region. We are working towards using Deep State Space Models (DSSM) based approach in predicting the energy demand vs supply trends. And this deep learning based technique has the potential to scale well when we transition to make predictions across U.S; while still maintaining the interpret-ability of the predictions.

Our code and report can be found here: https://github.com/adityakoolkarni/GenAutoNN_Reserve_Forecast