INTERACTIVE APPLICATIONS IN MACHINE LEARNING

14
Atau Tanaka - [email protected] Francisco Bernardo - [email protected] INTERACTIVE APPLICATIONS IN MACHINE LEARNING WORKSHOP Musical Gesture as Creative Interface Conference Universidade Católica Portuguesa 2016

Transcript of INTERACTIVE APPLICATIONS IN MACHINE LEARNING

Page 1: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

Atau Tanaka - [email protected] Francisco Bernardo - [email protected]

INTERACTIVE APPLICATIONS IN MACHINE LEARNINGWORKSHOP

Musical Gesture as Creative Interface Conference

Universidade Católica Portuguesa

2016

Page 2: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

BRIEFING

• Presentation of materials• Hardware - Bitalino Dev Kit • Software - EMG Max patch, Wekinator, Blotar

• Walkthrough for the workshop• Installation • Overview of EMG signal conditioning in Max • Sonification I • Regression with Machine Learning • Sonification II - Blotar~ • Free activity

Page 3: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

MATERIALS

• Hardware:• Bitalino Revolution Kit (x 10) • Electrodes (x 3) • Cables (x 1)

• Software (folder in USB flash) • Bitalino Max External Object • Max Patch with LFP/RMS & abstractions • Wekinator 2.0.1

Page 4: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Pair Bitalino with your computer• Go to Bluetooth preferences on Mac OS X sys preferences • Turn-on Bluetooth; Turn on Bitalino; • Select Bitalino from list of available devices; • Confirm your Bitalino MAC address label; • Click Options, introduce pairing code “1234” • Check the state of Bitalino in BT devices list (paired, but

not connected). • Rename your Bitalino with your unique identifier

Page 5: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Bitalino Plug’n’Play• Check back of the Bitalino board for EMG • Plug the cable to EMG socket of Bitalino • Glue electrodes in the arm (show best available positions

of the arm) • Connect cables’ terminals to electrode

• Middle cable (white) is reference. Please attach to non-fleshy arm area)

• Red and black on muscle electrodes

Page 6: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Setting up Bitalino External object in Max• Make sure you have Max in 64bit mode (get info, switch off

32bit mode) • Start Max from Applications folder, not from the dock! • Connect (if problems with port - message shown in Max

terminal, restart Max or even computer) • Start Max audio • Check EMG signal on visualiser

Page 7: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Conditioning Bitalino EMG signal with Max• Bitalino connection and modes • Visualising the signal

• Data for sonification and control• Low Pass filter to smooth data

• Smoothing - ~0.97 • Post gain - ~4.0

• Calibration - Toggle button to get local min/max - relax/tense • Sonification

• Raw signal sonification • Out to MIDI (set mode Off -> Glissando -> )

Page 8: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

Page 9: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Machine Learning setup with Wekinator• Setting OSC connection to signal source • Setting inputs and outputs • Recording, training and running the models

• Connection to Blotar~• Overview of Blotar~ object and path • Sonification with Blotar~

Page 10: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Install Wekinator• Install Wekinator from USB flash drive • Watch for security settings with the install of Wekinator

2.0.1 • Lauch Wekinator,

• Check port 6448 • Click start listening • Set #inputs to 1, set #outputs to 9 • Press Next

Page 11: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

Page 12: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

• Interactive Machine Learning loop with Wekinator• Verify OSC inputs are the expected (green status) • Record gesture #1 and map parameters

• Press Preset #1 on Blotar patch • Check changes on Wekinator’s Output parameters • Relax arm, press Recording button • Press Stop button • Press Train button

• Record gesture #2 do the same with tense arm • Press Run for performance mode with new data

Page 13: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

WALKTHROUGH

Page 14: INTERACTIVE APPLICATIONS IN MACHINE LEARNING

Atau Tanaka - [email protected] Francisco Bernardo - [email protected]

INTERACTIVE APPLICATIONS IN MACHINE LEARNINGWORKSHOP

Musical Gesture as Creative Interface Conference

Universidade Católica Portuguesa

2016

THANK YOU!