Below are some of my recent projects.

1. Intention Recognition from EEG Data
   Explored various CNN, RNN, Conv-Recurrent Net architectures in Keras. Implemented data prepossessing techniques like Windowed cropping, Sub-sampling to obtain the best performance of 67% with stacked GRU compared to a base line model of 46% with Shallow CNN.
   
   
2. SpaceInvaders using Deep RL Implemented Dueling, Double DQN to play Atari Space Invaders game in tensorflow.Dueling DQN learned better hiding strategy than Double DQN when trained for similar duration of time in both RAM and Pixel versions of the game. They achieved an average score of 351.1 and 291.1 respectively.
   
   
3. Path tracking using Deep RL Designed Deep RL based controller for Unicycle vehicle model using the DDPG algorithm to achieve continuous control in PyTorch.DDPG is more robust to action space noise than baseline LTV-MPC algorithm.
   
   
4. Movie Recommender system Implemented user-based K-NN, Latent factor based models like NNMF, SVD to predict movie rating in Movielens dataset. SVD performed better performance in terms of AUC.
   
   
5. Text Independent Automatic Speech Recognition in Noisy conditions I have investigated the effects using MFCC's, PNCC's, LPCC's, Rasta-PLP as feature vectors in noisy conditions. Experimented with filtering techniques like Spectral Subraction, MMSE to remove the noise. The best performance is achieved by PNCC's.