Artificial intelligence (AI) is being used in all disciplines in research and industry. AI models are influencing our lives in many areas (movies to watch, articles to read, music to listen, etc.). It is important to develop methods that allow us to understand the output produced by an AI model and build trust. Explainable AI field incorporates some of these methods.

In this workshop we will be using a method named Grad-CAM to help explain how AI models make their predictions. We will be detecting the type of scenery from the satellite images, visualize the Grad-CAM, and deploy the trained model using Tensorflow Serving.

Objectives:

Learn about what is Explainable AI and some of its techniques
Understand the theory and intuition behind Deep Neural Networks, Residual Nets, and Convolutional Neural Networks (CNNs)
Apply Python libraries to import, pre-process and visualize images
Perform data augmentation to improve model generalization capability
Build a deep learning model based on Convolutional Neural Network and Residual blocks (ResNet-18) using Keras with Tensorflow 2.0 as a backend
Compile and train the deep learning model
Assess the performance of trained CNN and evaluate its accuracy, precision and recall
Understand the theory and intuition behind GradCam and Explainable AI
Visualize the Activation Maps used by CNN to make predictions using Grad-CAM
Save the model for deployment and serve it using Tensorflow Serving

Level: Intermediate

Duration: 2 hours

Prerequisites: Prior experience with training models in Keras and Tensorflow is preferred but not required

Git repository: https://git.rcc.uchicago.edu/tszasz/explainable_ai

Interpreting Deep Learning Models Using Explainable AI