Graph regularization for image classification using synthesized graphs

Sayak_Paul · September 14, 2021, 1:28am

With sincere guidance from Arjun Gopalan from the Neural Structured Learning team, we just published a tutorial on using structural similarity to regularize the training of deep neural networks for image data:

github.com

tensorflow/neural-structured-learning/blob/master/neural_structured_learning/examples/notebooks/graph_keras_cnn_flowers.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "24gYiJcWNlpA"
      },
      "source": [
        "##### Copyright 2021 Google LLC"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "cellView": "form",
        "id": "ioaprt5q5US7"
      },
      "outputs": [],
      "source": [

This file has been truncated. show original

Under the hood, we represent the images of a given dataset as a synthesized graph wherein the nodes are the neighbors for a given image and the edges denote their similarity.

Works with Keras and introduces minimal technical overhead.