https://github.com/ashishbodhankar/mnist_classification_tensorflow_no_keras

Understanding the performance of different neural network architectures on the MNIST handwritten digits dataset, implemented in Tensorflow.
https://github.com/ashishbodhankar/mnist_classification_tensorflow_no_keras

categorical-cross-entropy cnn custom-tensorflow gradienttape mnist mnist-handwriting-recognition regularization sequential-models tensorflow tensorflow-models

Last synced: 6 months ago
JSON representation

Understanding the performance of different neural network architectures on the MNIST handwritten digits dataset, implemented in Tensorflow.

• Host: GitHub
• URL: https://github.com/ashishbodhankar/mnist_classification_tensorflow_no_keras
• Owner: AshishBodhankar
• Created: 2021-12-07T05:56:37.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2021-12-13T09:21:19.000Z (over 4 years ago)
• Last Synced: 2025-02-13T18:49:31.561Z (over 1 year ago)
• Topics: categorical-cross-entropy, cnn, custom-tensorflow, gradienttape, mnist, mnist-handwriting-recognition, regularization, sequential-models, tensorflow, tensorflow-models
• Language: Jupyter Notebook
• Homepage:
• Size: 237 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# TensorFlow framework (without Keras) to develop a multi-class classification problem on the MNIST handwritten digits dataset.

1. We built the architecture by custom defining the neural network class. Here, we use GradientTape to record operations for automatic differentiation. Firstly, we consider building a fully connected neural network with 2 hidden layers (no regularization) and measure the performance.

We also separately have written another (second) script for the same, but, this time including the Keras commands and, -

	1. We build the same network as before.

	2. We then add regularization and compare the difference between L2 and Dropout.

	3. lastly, we add a few CNN layers and see if we can improve the model's performance further.