Activation Bottleneck: Sigmoidal Neural Networks Cannot Forecast a Straight Line

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Toller, Maximilian, Hussain, Hussain, Geiger, Bernhard C
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913376952647680
author Toller, Maximilian
Hussain, Hussain
Geiger, Bernhard C
author_facet Toller, Maximilian
Hussain, Hussain
Geiger, Bernhard C
contents A neural network has an activation bottleneck if one of its hidden layers has a bounded image. We show that networks with an activation bottleneck cannot forecast unbounded sequences such as straight lines, random walks, or any sequence with a trend: The difference between prediction and ground truth becomes arbitrary large, regardless of the training procedure. Widely-used neural network architectures such as LSTM and GRU suffer from this limitation. In our analysis, we characterize activation bottlenecks and explain why they prevent sigmoidal networks from learning unbounded sequences. We experimentally validate our findings and discuss modifications to network architectures which mitigate the effects of activation bottlenecks.
format Preprint
id arxiv_https___arxiv_org_abs_2406_02146
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Activation Bottleneck: Sigmoidal Neural Networks Cannot Forecast a Straight Line
Toller, Maximilian
Hussain, Hussain
Geiger, Bernhard C
Machine Learning
A neural network has an activation bottleneck if one of its hidden layers has a bounded image. We show that networks with an activation bottleneck cannot forecast unbounded sequences such as straight lines, random walks, or any sequence with a trend: The difference between prediction and ground truth becomes arbitrary large, regardless of the training procedure. Widely-used neural network architectures such as LSTM and GRU suffer from this limitation. In our analysis, we characterize activation bottlenecks and explain why they prevent sigmoidal networks from learning unbounded sequences. We experimentally validate our findings and discuss modifications to network architectures which mitigate the effects of activation bottlenecks.
title Activation Bottleneck: Sigmoidal Neural Networks Cannot Forecast a Straight Line
topic Machine Learning
url https://arxiv.org/abs/2406.02146