The main language for machine learning has been Python for quite a while and the platform for training models has been servers, preferably with many powerful GPUs.
However, a browser running JavaScript is especially interesting when you want to visualize results or intermediate steps of machine learning processes in a truly interactive way. This can be both interesting for education as well as for debugging your models.
The browser can also be a platform for making interferences or predictions based on models that have been trained offline by much more powerful machines. This can be helpful to bring machine learning to a zero installation environment and thus closer to potential non technical user.
In this talk I will introduce you to the TensorFlow Playground that helps to understand how machine learning with deep neural networks works. To better understand how specific neural networks make their inferences, there are also some great browser based visualizations, among them ConvNet.js and tsne.js. We will also have a look at them. Finally, you will see how to make inferences in the browser using real Keras based models that can even make use of GPUs.
Required audience experience: Basic machine learning and JavaScript knowledge.
Objective of the talk: Learn about JavaScript and Machine Learning in the browser.
Keywords: ConvNet.js, tsne.js, Machine Learning, Browser, JavaScript, Keras, TensorFlow
You can view Oliver’s slides here:
https://djcordhose.github.io/ai/2017_ml_browser.html#/
You can view Oliver’s presentation here: