How Does Visualization Help People Learn Deep Learning? Evaluating GAN Lab with Observational Study and Log Analysis

Fast forward

Direct link to video on YouTube: https://youtu.be/SC04qeh1nWc

Keywords

Abstract

While a rapidly growing number of people want to learn artificial intelligence (AI) and deep learning, the increasing complexity of such models poses significant learning barriers. Recently, interactive visualizations, such as TensorFlow Playground and GAN Lab, have demonstrated success in lowering these barriers. However, there has been little work in evaluating these tools with human subjects. This paper presents two studies on evaluating GAN Lab, an interactive tool designed to help people learn how Generated Adversarial Networks (GANs) work. First, through an observational study, we investigate how the tool is used and what users learn from their usage. Second, we conduct a log analysis of the deployed tool to investigate how its visitors engage with GAN Lab. Based on the studies and our experience in developing and successfully deploying the tool, we provide design considerations and discuss further evaluation challenges for interactive educational tools for deep learning.