Every year the Glowacki group does a small hackathon. Since our group spans such a wide range of backgrounds and research interests, this yearly event attempts to expose all members of the group to some useful computational tools. Last year for example, Mike taught us the basics of Unity and we all had a go at making some simple games in VR.

This year, it was my turn to organise the hackathon. I decided to focus it on the software library “TensorFlow”, which is one of the most commonly used software libraries for implementing machine learning algorithms.

The hackathon was divided into two parts. The first part was a tutorial where I introduced the basics of TensorFlow while the second part was more practical. The tutorial included some simple examples going from linear regression to using small neural networks to fit cubic functions.

For example, below are some pictures obtained with one of the scripts from the tutorial. On the left you can see a (over)fit of a cubic function obtained with a small neural network with only one hidden layer. The diagram on the right is a visualisation of the data flow graph used to perform the fit of the cubic function.

Cubic function fit by a neural network

Data flow graph

If you are curious about TensorFlow, the slides and the coded examples for the tutorial part of the hackathon can be found on my github here, and in particular the script used to generate the pictures above is here.

### Like this:

Like Loading...

*Related*

Reblogged this on Glowacki Research Group.

LikeLike