This repository has the assignments I completed as part of my Deep Learning course for my Master's in Data Analytics program.
MNIST Neural Networks - Binary and Multi-Class Classification
Developed Artificial Neural Networks (ANN) for Binary and Multi-Class Classification using the MNIST digit dataset. For Binary Classification, trained models to distinguish selected digits, explore different weight initializers, and evaluate performance with confusion matrices and learning curves. In Multi-Class Classification, built a comprehensive model considering all 10 classes, incorporating early stopping and presenting a detailed confusion matrix analysis.
CNN and ResNet Model Implementation and Evaluation
Implemented a 5-layer CNN with Batch Normalization, MaxPooling, and 5-fold cross-validation with Grid Search for hyperparameter tuning on the Fashion MNIST dataset. Explored data augmentation, and transfer learning with VGG-19, and developed a custom ResNet model, employing various deep learning techniques for computer vision.
Variational Autoencoder for Fashion MNIST and Sentiment Analysis on IMDB
Implemented a Variational Autoencoder for Fashion MNIST, visualizing latent space clustering, and conducted sentiment analysis on IMDB Movie Reviews using TF-IDF + GaussianNB, Word2Vec + GaussianNB, and Glove + GaussianNB models. Byte Pair Encoding (BPE) to varying percentages (5%, 10%, 15%, 20%) of the IMDB training dataset, comparing token counts.
Sentiment Analysis and Text Generation with Embeddings
Performed sentiment analysis on the IMDB Movie Review dataset, exploring various embeddings and sequential models. Utilized the Keras embedding layer for sentiment analysis and applied cosine similarity to identify words similar to "movie." Executed text generation on the Moby Dick Chapter Four dataset with Keras word embeddings. Implemented transfer learning using Word2Vec embeddings.