Classifying the classifier: dissecting the weight space of neural networks

Eilertsen, Gabriel,1984-Linköpings universitet,Medie- och Informationsteknik,Tekniska fakulteten (författare)

Classifying the classifier : dissecting the weight space of neural networks

Funding: Wallenberg Autonomous Systems and Software Program (WASP); strategic research environment ELLIIT
This paper presents an empirical study on the weights of neural networks, where we interpret each model as a point in a high-dimensional space – the neural weight space. To explore the complex structure of this space, we sample from a diverse selection of training variations (dataset, optimization procedure, architecture,etc.) of neural network classifiers, and train a large number of models to represent the weight space. Then, we use a machine learning approach for analyzing and extracting information from this space. Most centrally, we train a number of novel deep meta-classifiers withthe objective of classifying different properties of the training setup by identifying their footprints in the weight space. Thus, the meta-classifiers probe for patterns induced by hyper-parameters, so that we can quantify how much, where, and when these are encoded through the optimization process. This provides a novel and complementary view for explainable AI, and we show how meta-classifiers can reveal a great deal of information about the training setup and optimization, by only considering a small subset of randomly selected consecutive weights. To promote further research on the weight space, we release the neural weight space (NWS) dataset – a collection of 320K weightsnapshots from 16K individually trained deep neural networks.

Jönsson, Daniel,1984-Linköpings universitet,Medie- och Informationsteknik,Tekniska fakulteten(Swepub:liu)danjo37 (författare)
Ropinski, TimoInstitute of Media Informatics, Ulm University, Germany (författare)
Unger, Jonas,1978-Linköpings universitet,Medie- och Informationsteknik,Tekniska fakulteten(Swepub:liu)jonun48 (författare)
Ynnerman, Anders,1963-Linköpings universitet,Medie- och Informationsteknik,Tekniska fakulteten(Swepub:liu)andyn27 (författare)
Linköpings universitetMedie- och Informationsteknik (creator_code:org_t)

Ingår i:Proceedings of the 24th European Conference on Artificial Intelligence (ECAI 2020): IOS PRESS, s. 1119-11269781643681016

Av författaren/redakt...: Eilertsen, Gabri ...; Jönsson, Daniel, ...; Ropinski, Timo; Unger, Jonas, 19 ...; Ynnerman, Anders ...

Om ämnet

NATURVETENSKAP: NATURVETENSKAP; och Data och informa ...; och Datorseende och ...

Kungliga biblioteket hanterar dina personuppgifter i enlighet med EU:s dataskyddsförordning (2018), GDPR. Läs mer om hur det funkar här.
Så här hanterar KB dina uppgifter vid användning av denna tjänst.