Recap from Day 020

In day 020, we looked at common soft clustering algorithms. We saw that Fuzzy clustering is a form of clustering in which each data point can belong to more than one cluster, while Gaussian mixture model is a probabilistic model that assumes all the data points are generated from a mixture of a finite number of Gaussian distributions with unknown parameters.

Using Unsupervised Learning To Come Up With New Features.

In a previous article, we talked about unsupervised learning and how it can be used. Today, in the words of Dr Rebecca Fiebrink, let’s see how unsupervised learning can be used for coming up with new features.

A simple scenario; You are working with a dancer and capturing the movement of the dancer with a Kinect, perhaps you start by using a simple feature representation, the X, Y and Z position of each of the dancers joint in space. You could use an unsupervised clustering techniques to discover that maybe there are five types of movement that the dancer often uses, and they are somewhat distinct in the feature space.

Source

Having discovered what the movements are, you might then write a computer program to watch out for them during a performance, and then change the lighting or the visuals when each category of movement is detected, or you might do something more interesting. You could use the category membership of the dancer’s current movement as a feature.

Source

You can assume that that feature tells you something meaningful about the dancer. Maybe it’s something about the dancer’s movement quality or where the dancer is in the structure of a piece, and you could use that feature as an input to a classifier or a regression algorithm in order to create a more complex analysis or a control system.

Great Job. You made it to the end of day 021. I hope you found this informative. Thank you for taking time out of your schedule and allowing me to be your guide on this journey.

Reference

https://www.kadenze.com/courses/machine-learning-for-musicians-and-artists-v/sessions/developing-a-practice-with-machine-learning-wrap-up