• Home
  • Search
  • Approximate Inverse Model Explanations as a Model-Agnostic Framework for Linking Musical Scores and Performance
  • https://doi.org/10.1109/access.2026.3673347Copy DOI Icon

Approximate Inverse Model Explanations as a Model-Agnostic Framework for Linking Musical Scores and Performance

Show More
  • Abstract
  • Literature Map
  • References
  • Similar Papers
Abstract

Understanding how written musical structures relate to expressive timing decisions is central to computational performance analysis; however, existing explanation methods often rely on predictive models, surrogate learning, or input perturbation. This study proposes approximate inverse model explanations (AIME) as a model-agnostic and data-efficient framework for analyzing expressive piano performance. Unlike perturbation-based approaches such as Shapley additive explanations and local interpretable model-agnostic explanations, AIME derives explanatory contributions directly from observed data by applying a Moore–Penrose inverse to matrices of score-derived features <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">(X)</i> and expressive events <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">(Y)</i> without requiring model training or sampling. Using the Vienna 4×22 corpus, with additional validation on the heterogeneous ASAP dataset, we project score landmarks onto a beat-synchronous grid and extract sparse expressive change points (knees) from MIDI performances. AIME computes the importance of global features (GFI<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><sup>corpus</sup></i>) and the importance of local features (LFI) within a unified algebraic framework. Experimental results show that boundary-related features, particularly slurs and barlines, consistently dominate structural importance, whereas effect-size analysis indicates a suppression of expressive events near phrase boundaries. Although symbolic structural features alone do not strongly separate individual performers, robustness diagnostics confirm stable and reproducible explanations across datasets. These findings demonstrate that AIME provides transparent, structural interpretations of expressive timing and generalizes to small, heterogeneous performance corpora, establishing a practical explanation framework for symbolic music analysis without reliance on predictive modeling.

Similar Papers
  • Research Article
  • Citations49

Interpretation and Classification of Arrhythmia Using Deep Convolutional Network

  • Jan 01, 2022
  • IEEE Transactions on Instrumentation and Measurement
  • Prateek Singh +1
  • Research Article
  • Citations22

Unlocking Machine Learning Model Decisions: A Comparative Analysis of LIME and SHAP for Enhanced Interpretability

  • Mar 31, 2024
  • Journal of Electrical Systems
  • Deepak Mane, Anand Magar, Om Khode, Sarvesh Koli, Komal Bhat, Prajwal Korade
  • Research Article

Explainable soft-voting classifier for heart disease prediction using SHAP and LIME

  • Feb 10, 2026
  • Discover Computing
  • Samiksha Walia +5
  • PDF
  • Research Article
  • Citations15

Machine learning-enabled prediction of prolonged length of stay in hospital after surgery for tuberculosis spondylitis patients with unbalanced data: a novel approach using explainable artificial intelligence (XAI)

  • Jul 25, 2024
  • European Journal of Medical Research
  • Parhat Yasin +6
  • PDF
  • Research Article
  • Citations4

Explainable Machine Learning to Predict Successful Weaning of Mechanical Ventilation in Critically Ill Patients Requiring Hemodialysis

  • Mar 21, 2023
  • Healthcare
  • Ming-Yen Lin +3
  • Research Article

Demystifying Green Innovation Impact on Firm Performance in Pakistan: A Comparative Analysis of SHAP and LIME Explainable AI Methods

  • Dec 11, 2025
  • Inge CUC
  • Shah Mehmood Wagan +1
  • Research Article

Hybrid BiLSTM-CNN Model with Attention and XAI (SHAP &amp; LIME) for ECG Arrhythmia Classification Using PTB-XL

  • Dec 31, 2025
  • International Journal of Science and Research (IJSR)
  • Maryam Shadan +4
  • Conference Article
  • Citations7

Determining the Major Contributing Features to Predict Breast Cancer Imposing ML Algorithms with LIME and SHAP

  • Nov 16, 2023
  • Prakash Paudel +3
  • Research Article
  • Citations13

Explainable artificial intelligence for predicting dengue outbreaks in Bangladesh using eco-climatic triggers.

  • Dec 01, 2025
  • Global epidemiology
  • Md Siddikur Rahman +1
  • Research Article

Development of a Hybrid Machine Learning Model for Predictive Performance Optimization in Lean Manufacturing and Industry 4.0

  • Dec 01, 2025
  • Sayeda Sufia Sumi
  • Research Article
  • Citations2

Improving Hepatitis B outcome prediction with ensemble machine learning: A study on predictive models and interpretability

  • May 01, 2025
  • Digital Health
  • Abid Bin Ahosan +4
  • Research Article
  • Citations13

Explainable artificial intelligence-machine learning models to estimate overall scores in tertiary preparatory general science course

  • Nov 20, 2024
  • Computers and Education: Artificial Intelligence
  • Sujan Ghimire +9
  • Research Article
  • Citations5

Application of an interpretable machine learning method to predict the risk of death during hospitalization in patients with acute myocardial infarction combined with diabetes mellitus

  • Apr 07, 2025
  • Acta Cardiologica
  • Zhijun Bu +12
  • Research Article
  • Citations3

Forecasting and Feature Analysis of Ship Fuel Consumption by Explainable Machine Learning Approaches

  • Mar 01, 2025
  • Polish Maritime Research
  • Nguyen Dang Khoa Pham +6
  • Conference Article

An Explainable AI-based Network Intrusion Detection System for Botnet Attacks

  • Jun 17, 2025
  • Dorieh Alomari +2
Cactus Communications logo

Copyright 2026 Cactus Communications. All rights reserved.