Machine Learning and Keras
a talk by Petr Baudiš
What is needed for machines to automatically teach themselves to make decisions based on examples? Machine Learning is all the rage nowadays and may seem very magical, especially on “unstructured data” like texts, pictures or sounds.
But what is actually a “neural network” and do you need a PhD to build some? Turns out that machine learning models are far less magical than they seem to be (at least when solving the more common tasks). Even if your program will acquire magic powers in the eyes of your users.
We will unveil the dark secret of how neural networks actually look inside. And to solve two real problems (related to text and pictures), we will actually build a couple of simple ones in the popular Python framework Keras.
This talk is suitable for both beginner and advanced Pythonistas.
I built machine learning based technologies for 10 years now, and build deep learning models for past couple of years.
Some of the interesting work I did is the Pachi Go program (used to be the best open source Go-playing program for many years) or the YodaQA open source question answering engine. I'm also Keras contributor.
I co-founded and currently lead the R&D at Rossum.ai, a startup building Artificial Intelligence for document understanding.