ApproxNet is a video analytics system for the edge. It enables novel dynamic approximation techniques to achieve desired inference latency and accuracy trade-off under different system conditions and resource contentions, variations in the complexity of the video contents and user requirements.
Study the population convergence of N-player game using learned strategies.
Study Adversarial attacks and Defence techniques for Machine Learning models.
A memory-incorporated RL framework that can learn to do Multiple tasks through active learning, and effectively reduce catastrophic forgetting on a set of Atari Games.
Study and evaluation of the performance Hierarchical Reinforcement Learning frameworks in multi-tasking domains using active sampling.
Implemented a table summarizer for structured weather data using an encoder-decoder model comprising of an attention layer over a hierarchical bidirectional LSTM based encoder and LSTM decoder.
Created a corpus and embeddings for nearly 15 million words in the Indian native language Malayalam.
We build a remote operated all-terrain rover, complete with a robotic manipulator and digger, with an in-built autonomous navigation module.