Projects

Machine Learning

• Implemented and analysed many ML algorithms:

• Data contest for predicting movie rating: Task was to predict movie rating, given dataset we had to try different collaborative filtering like- Nearest Neihjbour model. Got best with Modified latent factor model. With 73% accuracy, we achieved 11th rank on Kaggle.

Deep learning

• Compressed Representation of Data using Restricted Boltzmann Machine (RBM): Trained RBM using Contrastive Divergence (CD) algorithm to learn an n-dimensional hidden representation of 784-dimensional binary Fashion MNIST image dataset.

• Text Transliteration using LSTM based Encoder-Decoder: Performed English to Hindi Transliteration by training a bidirectional LSTM Encoder and 2-layered Decoder with attention mechanism with 52% accuracy.

• Image Classification using Convolutional Neural Network (CNN): Built a CNN using TensorFlow and trained it on a subset of ImageNet dataset for the classification of the images.

Reinforcement Learning

• Bandits: Implemented and did the comparative analysis for the following bandit algorithms- Epsilon-greedy, soft-max, UCB1, Median.

• RL algorithms comparative analysis: Implemented SARSA, Q-Learning, Sarsa Lambda which learns Q-values over states. and implemented policy gradient which directly learns policy over states and did the comparative analysis.

• Four rooms and the cart-pole: Implemented SMDP-Q learning and Intra-option learning on Four room environment and found that Intra-option learning is sample efficient. Tried DQN for cart pole and successfully completed the task of an average 195 reward over 100 episodes in less than 200 episodes.

  Other Projects -

• Developed Website of Literary Club (The Editorial Board) of MMMUT.

• Developed Web app Lets meet. Sep’ 2018 – Oct’ 2018