The Randomness of Input Data Spaces is an A Priori Predictor for Generalization

Publikationstyp:
Zeitschriftenaufsatz
Metadaten:
Autoren
Zeitschrift
CoRR
Datum der Veröffentlichung
2021
Titel
The Randomness of Input Data Spaces is an A Priori Predictor for Generalization.
Ausgabe der Zeitschrift
abs/2106.04181

Datenquelle: DBLP

Andere Metadatenquellen:
Abstract
Over-parameterized models can perfectly learn various types of data distributions, however, generalization error is usually lower for real data in comparison to artificial data. This suggests that the properties of data distributions have an impact on generalization capability. This work focuses on the search space defined by the input data and assumes that the correlation between labels of neighboring input values influences generalization. If correlation is low, the randomness of the input data space is high leading to high generalization error. We suggest to measure the randomness of an input data space using Maurer's universal. Results for synthetic classification tasks and common image classification benchmarks (MNIST, CIFAR10, and Microsoft's cats vs. dogs data set) find a high correlation between the randomness of input data spaces and the generalization error of deep neural networks for binary classification problems.
Autoren
Autoren-URL
http://arxiv.org/abs/2106.04181v2
Schlüsselwörter
  • cs.LG
  • cs.LG
  • cs.NE
Datum der Veröffentlichung
2021
Datum der Datenerfassung
2021
Datum, an dem der Datensatz öffentlich gemacht wurde
2021
Titel
The Randomness of Input Data Spaces is an A Priori Predictor for Generalization

Files

2106.04181v2.pdf

Datenquelle: arXiv

Autoren
Datum der Veröffentlichung
2021
Herausgeber URL
https://arxiv.org/abs/2106.04181
Datum der Datenerfassung
2021
Titel
The Randomness of Input Data Spaces is an A Priori Predictor for Generalization

Datenquelle: Manual

Beziehungen:
Eigentum von

Werkzeuge