Dueling Double DQN based Autonomous Vehicle Overtaking
Implement and Trained a ueling Double DQN (Deep Q-Learning Network) to overtake cars in a 3 lane scenario on highways.
3D Super-Resolution of CT and MRI Scans using CNN
Implemented and Trained the CNN-IL algorithm in the existing IEEE journal paper on IXI dataset. Proposed and Trained improved custom networks for super-resolution in height-and-width and depth directions in PyTorch by rewiring skip connections, adding ELU activations and adding more layers. Improved SSIM and PSNR using the above custom network.
Handwritten Digit Recognition
Implemented Logistic Regression, Kernel SVM with PCA, MDA and LeNet-5 CNN from scratch to identify/classify handwritten digits.
Face recognition
Implemented advanced machine learning algorithms, including Bayes' Classifier, kNN Classifier, Boosted SVM, and kernel SVM (using linear, polynomial, and RBF kernel), for face recognition without using in-built functions. Applied PCA to the dataset to find directions that maximize the variance and MDA to find the directions that maximize the separation between classes to further improve the prediction accuracy.
Fine grain classification using Transfer Learning
Implemented transfer learning using VGG to classify among 10 different species of monkeys, thus increasing the accuracy from 20.18% to 83.09%.