Neural Networks Use Distance Metrics

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autore principale: Oursland, Alan
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916497073373184
author Oursland, Alan
author_facet Oursland, Alan
contents We present empirical evidence that neural networks with ReLU and Absolute Value activations learn distance-based representations. We independently manipulate both distance and intensity properties of internal activations in trained models, finding that both architectures are highly sensitive to small distance-based perturbations while maintaining robust performance under large intensity-based perturbations. These findings challenge the prevailing intensity-based interpretation of neural network activations and offer new insights into their learning and decision-making processes.
format Preprint
id arxiv_https___arxiv_org_abs_2411_17932
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Neural Networks Use Distance Metrics
Oursland, Alan
Machine Learning
Artificial Intelligence
We present empirical evidence that neural networks with ReLU and Absolute Value activations learn distance-based representations. We independently manipulate both distance and intensity properties of internal activations in trained models, finding that both architectures are highly sensitive to small distance-based perturbations while maintaining robust performance under large intensity-based perturbations. These findings challenge the prevailing intensity-based interpretation of neural network activations and offer new insights into their learning and decision-making processes.
title Neural Networks Use Distance Metrics
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2411.17932