Fooling the Machine

doodle collection by a ML model

Nature of Code Final Project, ITP, New York

April 2017

Fooling the Machine was the final project of Nature of Code, given by prof. Daniel Shiffman at ITP, NYU.

In the class, I was fascinated by machine learning models for recognizing images after being trained. However, there are also people afraid of the idea, questioning whether machines will take over human beings one day. This project was made as a response to questions like this.

As the audience, you can draw anything on the tablet. No matter what you draw, the machine learning model will categorize the drawing as a hand-written digit and display it on the screen to form the outline of the digit, presenting a collection of what it thinks as a digit.

11.PNG

The model was trained on MNIST dataset so it can only recognize an image as one of ten digits. The project is meant to tell the audience that machine learning models are tools for us, so that we should train them carefully and take full advantage of them.

MnistExamples.png

The project was refined in April 2018 and was exhibited in Infinite exhibiton, held by MixCity in Shanghai

a1.jpeg
a2.jpeg