Embodied Machine Learning
An educational game to acquire a basic understanding of neural networks
Embodied machine learning is an educational game aimed at people trying to acquire a basic understanding of neural networks before moving on to more in-depth learning.
The game refrains from involving math and calculations and tries to create understanding through metaphors, embodiment, and computational thinking. Reducing complexity requires blurring concepts or omitting them altogether. I try to strike a balance between conceptual accuracy and accessibility; not to over-simplify, but also not to complicate.
In technical terms, the game is a (human) handwritten digit classifier trained on the MNIST dataset through supervised learning. Essentially, this means: a group of people together will be able to identify handwritten digits.
Identifying handwritten digits of course comes natural to most literate people (unless, perhaps, the digit is written by a doctor) but used to be quite hard a problem for computers. While the problem is long solved MNIST, a dataset of 60.000 labelled images of handwritten digits, remains a classic dataset and benchmarking tool for ML applications. I choose to emulate a MNIST classifier for this historic significance and to create a sense of purpose or achievement for players in my game.
The game is currently in prototype status. Read more on Github.
Consulting: Fei Liu
University: Parsons School of Design, The New School, NY