Image classification on the CIFAR 10 Dataset using Support Vector Machines (SVMs), Fully Connected Neural Networks and Convolutional Neural Networks (CNNs). The files are organized as follows:

SVMs_Part1 -- Image Classification on the CIFAR-10 Dataset using Support Vector Machines. Different types of kernels are used including Linear Kernel, Polynomial Kernel and the Radial Basis Function (RBF) Kernel.

SVMs_Part2_PCA -- Image Classification on the CIFAR-10 Dataset using Support Vector Machines. Principal Component Analysis (PCA) is used for dimensionality reduction. The number of dimensions are chosen based on the cumulative explained variance, as shown below:

Test results for the two transformations are shown below:

CNNs_Part1 -- Three different types of models are used in this case.

1. A Fully Connected Neural Network (3 Layers).
2. A Fully Convolutional Neural Network (4 Layers).
3. A Hybrid Model containing both convolutional and fully-connected layers.

The results are tabulated below:

CNNs_Part2 -- A Python File for training the convolutional neural network on the GPU.

See "Final_Report.pdf" for an analysis of the efficacy of the various algorithms on the CIFAR-10 Image Classification task. Some of the tricks used during training (e.g. cyclic learning rates) are also discussed.