Machine Learning

This project aims to compare the performance of various models on a binary classification task to distinguish between images of apples and bananas. The models being evaluated include different configurations of the VGG architecture, as well as transfer learning approaches.

This project focuses on training a Multi-Layer Perceptron (MLP) on the MNIST dataset, evaluating its performance against Random Forest (RF) and Logistic Regression models. The analysis includes F1-score and confusion matrix metrics, alongside visualizations using t-SNE to understand the embeddings from the MLP’s hidden layers.

This project leverages Multi-Layer Perceptrons (MLP) for next character prediction. Using a rich dataset of texts from William Shakespeare, this implementation is designed for text generation. This involves training a model that involves an embedding layer and multiple fully connected layers.

This project explores different models to tackle this binary classification task on XOR Dataset. With 200 training instances and 200 test instances, the project showcases model performance and decision surfaces, offering insight into the strengths and weaknesses of each approach.

This project leverages Random Fourier Features (RFF) to tackle complex image processing related tasks. This includes innovative techniques for image completion, reconstruction, and super-resolution. It also provides an in-depth analysis of the performance of RFF in comparison to traditional methods in this specfic task. This project is also source of learning for working on projects under limited computational resources (GPU).

This project discovers how different types of gradient descent work. Our project compares standard and momentum-based methods, showcasing how momentum accelerates convergence and enhances stability. Dive into our Jupyter notebooks and explore the behavior of Full Batch and Stochastic Gradient Descent—both with and without momentum.

This project implements decision trees for both classification and regression tasks! A decision tree algorithm is a supervised learning algorithm that uses a tree-like structure to classify data or predict outcomes. This versatile tool supports various input-output configurations, offering accurate predictions and insightful visualizations for data-driven decisions.