Day 6 - Task A

Problem Statement

Image Classification for Fashion MNIST Dataset

Requirements

pip install tensorflow

pip install matplotlib

pip install numpy

Dataset

The dataset can be found under Tensorflow Dataset.

Approach

Loaded the Dataset.
Developed a DNN with BatchNorm(Faster Convergence) and Dropouts(Regularization Technique).
Used One Cycle LR as an LR Scheduler and Trained the Model.
Visualized the Kernels.

Results

Model Parameters: 791,428
Train Accuracy: 94.55%
Test Acccuracy: 93.12%

Output

Dataset Gallery

Train - Test Curve

Input Image for Kernel Visualization

Kernel Visualization Plotting the output of the first kernel