Advent of numerous incredible technologies holds a promise of an amazing future. I am fortunate to know many of these technologies and my goal here is to help incubate high quality projects that has profound impact on common people's everyday lives.
I am Harshal Bharatia, a junior at the University of Texas at Austin in Computer Science Turing Scholar Honors program, with double major in Math and minor in Robotics. My current research interests include various topics in Artificial Intelligence and Machine Learning.
Heating, Ventillation and Cooling (HVAC) operation constitutes 45-55% of house energy, due to inefficient conventional thermostats. Goal of the project is to create a cloud-based collaborative learning thermostat that optimizes HVAC operation and maximizes comfort and energy savings using machine learning.
Unlike the other thermostats which attempt to conserve energy by reducing the time you run the HVAC system while you are away, this thermostat uses a patented technology that also learns how best to operate the HVAC system so that it can conserve energy while HVAC runs when you are home.
It automatically learns the patterns of operations that offer maximum savings and comfort, and collaborates with other thermostats with similar house profiles to further maximize savings.
Checkout this cool project in its entirety here.
This project implements a novel autonomous approach to knowledge refinement wherein the system automatically determines what can be improved and trains itself to do so. It enables solutions with smartness that goes beyond initial training, when domain is expansive and high quality training may not be feasible for entire expanse of the domain, or when spontaneous mutations alter various aspects of the domain.
A novel collaborative approach with automatically built dynamic ensembles of models target improvisations, such as improvements based on past knowledge or trends reflected in the results, and overshadows sub-optimal performance of existing model in a portion of input feature space. A model selector automatically identifies an optimal model for a query using reinforcement learning. This alleviates need for complex monolithic models requiring extensive training, automatically adapt to changes in the domain without destabilizing the models, offer better control over query performance, and can guide exploration when exploitation becomes inadequate.
The efficacy of this approach was demonstrated esp. for areas where deep-learning failed to perform for following applications:
There are numerous additional potential applications for this paradigm such as coaching sidekick for personalized training for student, athlete, astronauts; combat sidekick to adapt to a soldier/special ops needs; smart cities to adapt to unforeseen weather/social phenomenons and space exploration satellites that adapt during long journey in unknown conditions.
Checkout this cool project in its entirety here.
This is an education project where my goal is to give an honest comprehensive overview of the many prominent fields under Artificial Intelligence and then go a bit deeper in some of these fields to explore the field using some sample applications.
Checkout this cool project in its entirety here.
