machine learning

Academic Research

Summary

As part of my research work at the Center for New Music and Audio Technology (CNMAT, UC Berkeley), I explored the use of machine learning to map complex input signals into pitch and spatial coordinates.

In these projects, I used the LIBSVM machine learning library. Parameter optimization and statistical goodness-of-fit analysis were performed in Matlab and Mathematica.

Skills

Matlab
Mathematica
Python
LIBSVM
PyTorch
Support Vector Machine Learning
Training Methods
Goodness-of-fit Measures

Support Vector Machine Learning for Gesture Signal Estimation

I implemented a novel design for a pressure-sensitive e-textile-based touch surface proposed by Adrian Freed, Research Director at CNMAT. Using two layers of piezo-resistive fabric separated by a boundary layer, the interface can detect touch and pressure for up to two simultaneous points. However, due to irregularities in the fabric materials, it is necessary to use machine learning to estimate the position and pressure of touch points.

First, a training dataset was collected by sampling the device response on a grid; then, sensor data recordings were made along trial trajectories. After optimizing the parameters of a support vector regression model, I demonstrated it was possible to correct the data into a normalized XY plane with low error. In addition, by examining the magnitude of support vector coefficients, I showed that it is possible to quantify the locations within the surface responsible for sensor non-linearity.

Citation

Schmeder, Andrew, and Adrian Freed. "Support Vector Machine Learning for Gesture Signal Estimation with a Piezo-Resistive Fabric Touch Surface." NIME. 2010.

Mapping Spectral Frames to Pitch

In this project, I used Support Vector Machine Learning to map FFT signal frames to pitch using a database of guitar-note onset recordings recorded with a 6-channel pickup.

Using Kernel Principle Components Analysis, I demonstrated how eigenvectors of the pitch data are automatically projected onto the geometry of a spiral shape in 3-dimensional space (pictured above).

Citation

Schmeder, Andrew. "Mapping Spectral Frames to Pitch with the Support Vector Machine." ICMC. 2004.

Page updated

Google Sites

Report abuse