Matrix Optimizations, or Moxie, Khushboo Choudhari
Focused on optimizing SpMV (Sparse Matrix Vector Multiplication) operations by introducing a new vector arrangement technique and leveraging the vector units, resulting in a 30.29% boost in performance, along with a theoretical analytical model to analyze various matrix configurations. Additionally, worked on optimizing the EM (Expectation Maximization) process using the vector units. Paper published at the 2024 International IEEE HiPC (High Performance Computing) conference in Bangalore.
Autoencoder Exploration, or OX, Sanchi Jadhav
Solver Exploration, or SOX, Ketaki Dharmadhikari
Filter Exploration, or FOX, Mayuri Lomate
Perceptrons and SVMs, Anushka Desai and Sharayu Kondobhairy
The project involved exploring the Perceptron Learning Algorithm, with a focus on its mathematical proof of convergence and visualization. The work extended to an in-depth analysis of Support Vector Machines (SVMs) and kernel methods, particularly polynomial and RBF kernels. The study included investigating their mathematical foundations and analyzing their behavior on various datasets.
CNN to DNN Translator, Nandini Kanawade
Study the relationship and impact between convolutional layers and dense layers. The aim is to understand how these two types of layers contribute to feature extraction and classification tasks and leverage their unique strengths to create a translator model that combines the feature extraction power of convolutional layers with the computational efficiency and simplicity of dense layers.