Quantifying PTM-detector geometric alignment for reconstruction-based anomalous sound detection

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Hoang, Nguyen-Duc
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901864136572928
author Hoang, Nguyen-Duc
author_facet Hoang, Nguyen-Duc
contents <div> <div># Quantifying PTM-Detector Geometric Alignment for Reconstruction-Based ASD</div> <br> <div>This repository contains the source code necessary to reproduce the experimental results presented in the paper "Quantifying PTM-detector geometric alignment for reconstruction-based anomalous sound detection".</div> <br> <div>The project investigates the effectiveness of 12 diverse Pre-trained Models (PTMs) for reconstruction-based Anomalous Sound Detection (ASD) using Autoencoders (AEs). It quantifies **representational alignment** using **debiased Centered Kernel Alignment (CKA)** and defines the empirical upper bound via a **supervised Late Fusion** XGBoost meta-learner.</div> </div> <p># Project Directory Structure</p> <p>This section details the critical output directories generated by the execution pipeline, which store cached features, trained models, and all analytical results (AUC scores, CKA matrices, and final plots) necessary to reproduce the findings of the paper.</p> <p>The overall structure is defined in the configuration files (e.g., `extract_ast.yaml` and `extract_late_fusion.yaml`):</p> <p>```<br>.<br>├── dataset         # MIMII Dataset (Input data structure)<br>├── cache           # Stores PTM embeddings (HDF5 files)<br>├── model           # Stores trained Autoencoders (.pth)<br>└── result          # Stores final YAML/CSV results<br>    └── results_fusion # Fusion summary and plots (CKA, Figures)<br>``</p> <p>#All code in code.zip .</p> <p>#All data is the result of running the code in cache.zip, model.zip, results.zip.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17489579
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Quantifying PTM-detector geometric alignment for reconstruction-based anomalous sound detection
Hoang, Nguyen-Duc
<div> <div># Quantifying PTM-Detector Geometric Alignment for Reconstruction-Based ASD</div> <br> <div>This repository contains the source code necessary to reproduce the experimental results presented in the paper "Quantifying PTM-detector geometric alignment for reconstruction-based anomalous sound detection".</div> <br> <div>The project investigates the effectiveness of 12 diverse Pre-trained Models (PTMs) for reconstruction-based Anomalous Sound Detection (ASD) using Autoencoders (AEs). It quantifies **representational alignment** using **debiased Centered Kernel Alignment (CKA)** and defines the empirical upper bound via a **supervised Late Fusion** XGBoost meta-learner.</div> </div> <p># Project Directory Structure</p> <p>This section details the critical output directories generated by the execution pipeline, which store cached features, trained models, and all analytical results (AUC scores, CKA matrices, and final plots) necessary to reproduce the findings of the paper.</p> <p>The overall structure is defined in the configuration files (e.g., `extract_ast.yaml` and `extract_late_fusion.yaml`):</p> <p>```<br>.<br>├── dataset         # MIMII Dataset (Input data structure)<br>├── cache           # Stores PTM embeddings (HDF5 files)<br>├── model           # Stores trained Autoencoders (.pth)<br>└── result          # Stores final YAML/CSV results<br>    └── results_fusion # Fusion summary and plots (CKA, Figures)<br>``</p> <p>#All code in code.zip .</p> <p>#All data is the result of running the code in cache.zip, model.zip, results.zip.</p>
title Quantifying PTM-detector geometric alignment for reconstruction-based anomalous sound detection
url https://doi.org/10.5281/zenodo.17489579