Visualisation of hidden Layers

This repository contains implementation of a sequential model on the MNIST dataset along with the plot and visualization of hidden layers.

Dataset

The MNIST database is a dataset of handwritten digits. It has 60,000 training samples, and 28,000 test samples. Each image is represented by 28x28 pixels, each containing a value 0 - 255 with its grayscale value.

Model

This is a sequential model with three dense layers and some dropouts to avoid overfitting. The summary can be viewed with the command model.summary() after running the model.

• The model performance can be improved upto 98% using more number of epochs. If your your hardware supports it, use upto 100 epochs.

Hidden layers Visualization

I've used the weights of my model to build a new model that is truncated at the layer I want to read. And then I used TSNE and Bokeh to extract and visualize the embeddings of hidden layer data.