On Pitfalls of $\textit{RemOve-And-Retrain}$: Data Processing Inequality Perspective

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Song, Junhwa, Cha, Keumgang, Seo, Junghoon
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917181127655424
author Song, Junhwa
Cha, Keumgang
Seo, Junghoon
author_facet Song, Junhwa
Cha, Keumgang
Seo, Junghoon
contents Approaches for appraising feature importance approximations, alternatively referred to as attribution methods, have been established across an extensive array of contexts. The development of resilient techniques for performance benchmarking constitutes a critical concern in the sphere of explainable deep learning. This study scrutinizes the dependability of the RemOve-And-Retrain (ROAR) procedure, which is prevalently employed for gauging the performance of feature importance estimates. The insights gleaned from our theoretical foundation and empirical investigations reveal that attributions containing lesser information about the decision function may yield superior results in ROAR benchmarks, contradicting the original intent of ROAR. This occurrence is similarly observed in the recently introduced variant RemOve-And-Debias (ROAD), and we posit a persistent pattern of blurriness bias in ROAR attribution metrics. Our findings serve as a warning against indiscriminate use on ROAR metrics.
format Preprint
id arxiv_https___arxiv_org_abs_2304_13836
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle On Pitfalls of $\textit{RemOve-And-Retrain}$: Data Processing Inequality Perspective
Song, Junhwa
Cha, Keumgang
Seo, Junghoon
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Methodology
Approaches for appraising feature importance approximations, alternatively referred to as attribution methods, have been established across an extensive array of contexts. The development of resilient techniques for performance benchmarking constitutes a critical concern in the sphere of explainable deep learning. This study scrutinizes the dependability of the RemOve-And-Retrain (ROAR) procedure, which is prevalently employed for gauging the performance of feature importance estimates. The insights gleaned from our theoretical foundation and empirical investigations reveal that attributions containing lesser information about the decision function may yield superior results in ROAR benchmarks, contradicting the original intent of ROAR. This occurrence is similarly observed in the recently introduced variant RemOve-And-Debias (ROAD), and we posit a persistent pattern of blurriness bias in ROAR attribution metrics. Our findings serve as a warning against indiscriminate use on ROAR metrics.
title On Pitfalls of $\textit{RemOve-And-Retrain}$: Data Processing Inequality Perspective
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Methodology
url https://arxiv.org/abs/2304.13836