Self-Supervised Representation Learning on Neural Network Weights for Model Characteristic Prediction

Item Type	Conference or Workshop Item (Paper)
Abstract	Self-Supervised Learning (SSL) has been shown to learn useful and information- preserving representations. Neural Networks (NNs) are widely applied, yet their weight space is still not fully understood. Therefore, we propose to use SSL to learn neural representations of the weights of populations of NNs. To that end, we introduce domain specific data augmentations and an adapted attention architecture. Our empirical evaluation demonstrates that self-supervised representation learning in this domain is able to recover diverse NN model characteristics. Further, we show that the proposed learned representations outperform prior work for predicting hyper-parameters, test accuracy, and generalization gap as well as transfer to out-of-distribution settings.
Authors	Schürholt, Konstantin; Kostadinov, Dimche & Borth, Damian
Research Team	AIML Lab
Language	English
Subjects	computer science
HSG Classification	contribution to scientific community
Date	9 November 2021
Publisher	Neural Information Processing Systems (NeurIPS)
Place of Publication	Sydney, Australia
Volume	35
Event Title	Neural Information Processing Systems (NeurIPS)
Contact Email Address	konstantin.schuerholt@unisg.ch
Depositing User	Ulrich Konstantin Schürholt
Date Deposited	02 Nov 2021 15:15
Last Modified	14 Nov 2021 12:54
URI:	https://www.alexandria.unisg.ch/publications/264718

Download

Text
neurips_2021.pdf - Published Version
Download (5MB)

Citation

Schürholt, Konstantin; Kostadinov, Dimche & Borth, Damian: Self-Supervised Representation Learning on Neural Network Weights for Model Characteristic Prediction. 2021. - Neural Information Processing Systems (NeurIPS).

Statistics

https://www.alexandria.unisg.ch/id/eprint/264718

Edit item