Towards Deep Interpretability (MUS-ROVER II): Learning Hierarchical Representations of Tonal Music

Authors: Haizi Yu, Lav R. Varshney

ICLR 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 6 EXPERIMENTS MUS-ROVER II’s main use case is to produce personalized syllabi that are roadmaps to learning the input style (customized paths to Mount Parnassus). [...] We illustrate two syllabi in Table 1, which compares the first ten (1-gram) rules in a faster (γ = 0.5) syllabus and a slower one (γ = 0.1)." and "Let’s take a closer look at the two syllabi in Table 1. The specifications (left) and hierarchies (right) of the four common rules are illustrated in Table 2.
Researcher Affiliation Academia Haizi Yu Department of Computer Science University of Illinois at Urbana-Champaign Urbana, IL 61801, USA haiziyu7@illinois.edu Lav R. Varshney Department of Electrical and Computer Engineering University of Illinois at Urbana-Champaign Urbana, IL 61801, USA varshney@illinois.edu
Pseudocode Yes A.3 ALGORITHM FOR CONCEPTUAL HIERARCHY Algorithm 1: Algorithm for computing the conceptual hierarchy
Open Source Code No The paper provides a link to a dataset ("Harmonia’s Music XML corpus of Bach’s chorales (https://harmonia.illiacsoftware.com/)") but does not explicitly state that the code for the methodology described in this paper is open-source or provide a link to it.
Open Datasets Yes We use the same dataset that comprises 370 C scores of Bach’s 4-part chorales." and "We thank Professor Heinrich Taube, President of Illiac Software, Inc., for providing Harmonia’s Music XML corpus of Bach’s chorales (https://harmonia.illiacsoftware.com/),
Dataset Splits No The paper describes a self-learning loop and uses a dataset for training but does not specify explicit training/validation/test splits (e.g., percentages or counts) or cross-validation setup.
Hardware Specification No The paper does not explicitly describe the hardware (e.g., specific GPU or CPU models, memory) used to run its experiments.
Software Dependencies No The paper makes a brief mention of 'python nan variable' but does not provide specific version numbers for software dependencies or libraries used in their implementation.
Experiment Setup Yes There are two hyper-parameters δ and γ in the optimization problem (3), whose detailed usage in syllabus customization will be discussed later in Sec. 6. At a high level, we often pre-select γ before the loop to express a user’s satisfaction level: a smaller γ signifies a meticulous user who is harder to satisfy; the threshold δ upper bounds the entropic difficulty of the rules, and is adaptively adjusted through the loop: it starts from a small value (easy rules first), and auto-increases whenever the feasible set of (3) is empty (gradually increases the difficulty when mastering the current level).