EDS Hackathon: Machine Learning with Signal Analysis Techniques

Dear data enthusiasts,

Recurrent Neural Networks are a really popular method for analyzing data in the form of signals / time-series. But did you know you can do the same thing with traditional techniques from the world of Signal and Image processing?

The Autocorrelation, Power Spectral Density, Fourier Transform and the Discrete Wavelet Transform can be used to generate features from signals or time-series. These features can then be used in traditional classification models like Logistic Regression of Random Forest.

One of the main disadvantages of Deep Learning techniques is that they are black box models. They work very well and are accurate, but you don't know why or how it works. That is not the case with these Signal Processing techniques; since you are constructing the features yourself, you know exactly what the classification is based on and this also gives you more insight into the nature of the signal.

Interesting articles:

Have We Forgotten about Geometry in Computer Vision? (https://alexgkendall.com/computer_vision/have_we_forgotten_about_geometry_in_computer_vision/)

Deep, Deep Trouble: Deep Learning’s Impact on Image Processing, Mathematics, and Humanity (https://sinews.siam.org/Details-Page/deep-deep-trouble)

Agenda:

18:00 Walk in
18:15 Meetup starts: discuss theory of signal analysis techniques
18:45 Start hacking on data set
20:45 Collect submissions, announce winner and wrap up
21:00-21:30 drinks

Location:
The location (http://www.pluginpaviljoen.nl/fotos/) has a bar which we host ourselves. Please take some cash money to pay for your drinks.

https://secure.meetupstatic.com/photos/event/4/5/b/0/600_460337840.jpeg

https://github.com/RobRomijnders/EDS/raw/master/hack_12/glaspaviljoen.png