Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generating Piano Practice Policy with a Gaussian Process

Alexandra Moringen, Elad Vromen, Helge Ritter, Jason Friedman

TL;DR

This work presents a computational architecture building on a Gaussian process that incorporates the learner state, a policy that selects a suitable practice mode, performance evaluation, and expert knowledge, and the proposed policy model is trained to approximate the expert-learner interaction during a practice session.

Abstract

A typical process of learning to play a piece on a piano consists of a progression through a series of practice units that focus on individual dimensions of the skill, the so-called practice modes. Practice modes in learning to play music comprise a particularly large set of possibilities, such as hand coordination, posture, articulation, ability to read a music score, correct timing or pitch, etc. Self-guided practice is known to be suboptimal, and a model that schedules optimal practice to maximize a learner's progress still does not exist. Because we each learn differently and there are many choices for possible piano practice tasks and methods, the set of practice modes should be dynamically adapted to the human learner, a process typically guided by a teacher. However, having a human teacher guide individual practice is not always feasible since it is time-consuming, expensive, and often unavailable. In this work, we present a modeling framework to guide the human learner through the learning process by choosing the practice modes generated by a policy model. To this end, we present a computational architecture building on a Gaussian process that incorporates 1) the learner state, 2) a policy that selects a suitable practice mode, 3) performance evaluation, and 4) expert knowledge. The proposed policy model is trained to approximate the expert-learner interaction during a practice session. In our future work, we will test different Bayesian optimization techniques, e.g., different acquisition functions, and evaluate their effect on the learning progress.

Generating Piano Practice Policy with a Gaussian Process

TL;DR

This work presents a computational architecture building on a Gaussian process that incorporates the learner state, a policy that selects a suitable practice mode, performance evaluation, and expert knowledge, and the proposed policy model is trained to approximate the expert-learner interaction during a practice session.

Abstract

A typical process of learning to play a piece on a piano consists of a progression through a series of practice units that focus on individual dimensions of the skill, the so-called practice modes. Practice modes in learning to play music comprise a particularly large set of possibilities, such as hand coordination, posture, articulation, ability to read a music score, correct timing or pitch, etc. Self-guided practice is known to be suboptimal, and a model that schedules optimal practice to maximize a learner's progress still does not exist. Because we each learn differently and there are many choices for possible piano practice tasks and methods, the set of practice modes should be dynamically adapted to the human learner, a process typically guided by a teacher. However, having a human teacher guide individual practice is not always feasible since it is time-consuming, expensive, and often unavailable. In this work, we present a modeling framework to guide the human learner through the learning process by choosing the practice modes generated by a policy model. To this end, we present a computational architecture building on a Gaussian process that incorporates 1) the learner state, 2) a policy that selects a suitable practice mode, 3) performance evaluation, and 4) expert knowledge. The proposed policy model is trained to approximate the expert-learner interaction during a practice session. In our future work, we will test different Bayesian optimization techniques, e.g., different acquisition functions, and evaluate their effect on the learning progress.
Paper Structure (17 sections, 5 equations, 3 figures)

This paper contains 17 sections, 5 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Experimental Setting
  3. Participants
  4. Equipment
  5. Selection of piece to play
  6. Practice Modes
  7. Teacher role
  8. Methods: Models and Features
  9. Calculation of features from user data
  10. Gaussian process and the scaffolding method
  11. Methods - Linear regression
  12. Results
  13. Gaussian process
  14. Linear Regression
  15. Discussion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: The experimental setup is shown in this photo. The learner sits at the piano (Nord piano 4) with a computer monitor showing the notes to play. Using the Python interface, the teacher selects which practice mode to play and when to change the piece.
  • Figure 2: The upper picture shows an example of one of the pieces to be played. The lower picture shows the selection of a timing practice mode (PM) - the pitch of all notes is uniform, allowing the learner to focus on the timing.
  • Figure 3: (Upper left) An illustration of data points recorded during expert-guided piano practice session for all learners and pieces. $x$-axis denotes timing error; $y$-axis denotes pitch error before a PM (pitch practice in violet, timing practice in yellow). (Upper right, lower panels) Policies generated by a GP RatQuad kernel, BMP=50, 80 and 100. For slower tempos, the model's policy predicts a higher utility of timing practice; for higher tempos - a higher utility of pitch practice.