Worked on various reinforcement learning algorithms used for training an agent to play Atari games. Learned about DQN, DDQN, Duelling-DQN, DDPG, A3C, A2C, ACKTR, World Models and other Actor-Critic algorithms. Analysed the algorithms to understand the differences in learning and time taken by an agent to learn. Incorporated various evolutionary strategies such as Differential Evolution (DE) and CMA-ES. Incorporated different model architectures such as DC-GAN BE-GAN, Cycle GAN and Siamese network on the existing RL algorithms to see the performance boost.  

The internship is focused on finding motifs in a time series using PM2.5 sensors data from sources like LASS, LASS4U, Indie, ProbeCube, WEBDUINO, AirBox, AsusAirBox, MAPS, EPA, and CWB. Analysing the two algorithms of detecting motifs using Matrix Profile in one and discretization and dimensionality reduction in other. Applied Co-relational analysis between the two motifs to find the relationship between the motifs a novel idea.