Skip to main content
Talk

Machine-learning demonstration and hands-on workshop — University of Manchester DigiLab

Dr. Omar Costilla Reyes led a machine-learning demonstration and programming session at the University of Manchester Library DigiLab on March 2, 2017. It was built around a Kinect motion-tracking demo.

A short preview shared the day before Dr. Costilla Reyes’s University of Manchester DigiLab workshop. The clip previews a machine-learning Kinect demonstration with salsa steps rather than documenting the full workshop. Watch on YouTube.
Hands-on machine-learning session in the University of Manchester Library DigiLab Learning Commons, March 2017, led by Dr. Costilla Reyes. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Visitors try the Kinect motion-tracking demonstration during the DigiLab session. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Attendees working through the hands-on programming part of the DigiLab session. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Motion-tracking output from the Kinect demonstration shown at DigiLab, with a colored skeleton drawn for each person in view. Credit: Image courtesy of Dr. Omar Costilla Reyes.

A machine-learning day at the Library DigiLab

The University of Manchester Library hosted a technology-filled DigiLab day at its Learning Commons on March 2, 2017. Dr. Omar Costilla Reyes, then a doctoral researcher at the university, ran a machine-learning interactive demonstration and hands-on programming session as part of it. The day before, he shared a short preview video of the Kinect motion-sensing setup, salsa steps included, and afterward he thanked attendees for a good turnout and offered to share the code and resources.

Learning from movement

A Kinect is a depth camera: instead of recording color alone, it measures how far each point in the scene is from the sensor, and software can use those distances to estimate where a person’s head, shoulders, elbows, hips, and knees are. The photographs on this page show the result as colored stick-figure skeletons drawn over several people at once. Body tracking of this kind is itself built with machine learning, and the movements it records, such as a dance step, are numerical data that a program can compare and classify. That makes movement a vivid way to show visitors how a computer learns from examples, and the session paired the live demonstration with hands-on programming. The photographs also show attendees working at laptops and a browser-based neural-network tool on a screen.

Connection to his doctoral research

The theme connected to his doctoral research on the university’s smart-carpet floor sensor, which records footsteps rather than body joints. He also demonstrated that system to the public through the library’s DigiLab activities and the university’s Open Labs, where the question was what machine learning can infer about the way you walk. The gait models from that research were published in “Temporal pattern recognition in gait activities recorded with a footprint imaging sensor system” (2016) and developed in his University of Manchester thesis on pattern recognition from raw spatio-temporal data. The DigiLab session was one of several public-engagement activities during his doctorate, alongside STEM Ambassador work with the Museum of Science and Industry and the university’s Industry 4.0 Summit stand in 2017.

Topics