Deep Neural Networks (DNNs) have achieved remarkable performance on a range of tasks. A key step to further empowering DNN-based approaches is improving their explainability. In this work we present CME: a concept-based model extraction framework, used for analysing DNN models via concept-based extracted models. Using two case studies (dSprites, and Caltech UCSD Birds), we demonstrate how CME can be used to (i) analyse the concept information learned by a DNN model (ii) analyse how a DNN uses this concept information when predicting output labels (iii) identify key concept information that can further improve DNN predictive performance (for one of the case studies, we showed how model accuracy can be improved by over 14%, using only 30% of the available concepts).

Now you see me (CME): Concept-based model extraction / Kazhdan, D.; Dimanov, B.; Jamnik, M.; Lio, P.; Weller, A.. - 2699:(2020). (Intervento presentato al convegno 2020 International Conference on Information and Knowledge Management Workshops, CIKMW 2020 tenutosi a Galway; irl).

Now you see me (CME): Concept-based model extraction

Lio P.
;
2020

Abstract

Deep Neural Networks (DNNs) have achieved remarkable performance on a range of tasks. A key step to further empowering DNN-based approaches is improving their explainability. In this work we present CME: a concept-based model extraction framework, used for analysing DNN models via concept-based extracted models. Using two case studies (dSprites, and Caltech UCSD Birds), we demonstrate how CME can be used to (i) analyse the concept information learned by a DNN model (ii) analyse how a DNN uses this concept information when predicting output labels (iii) identify key concept information that can further improve DNN predictive performance (for one of the case studies, we showed how model accuracy can be improved by over 14%, using only 30% of the available concepts).
2020
2020 International Conference on Information and Knowledge Management Workshops, CIKMW 2020
Concept extraction; Concept-based explanations; Interpretability; Latent space analysis; Model extraction; Xai
04 Pubblicazione in atti di convegno::04b Atto di convegno in volume
Now you see me (CME): Concept-based model extraction / Kazhdan, D.; Dimanov, B.; Jamnik, M.; Lio, P.; Weller, A.. - 2699:(2020). (Intervento presentato al convegno 2020 International Conference on Information and Knowledge Management Workshops, CIKMW 2020 tenutosi a Galway; irl).
File allegati a questo prodotto
File Dimensione Formato  
Kazhdan_Now-you_2020.pdf

accesso aperto

Note: https://ceur-ws.org/Vol-2699/paper02.pdf
Tipologia: Versione editoriale (versione pubblicata con il layout dell'editore)
Licenza: Creative commons
Dimensione 8.41 MB
Formato Adobe PDF
8.41 MB Adobe PDF

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11573/1720250
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 10
  • ???jsp.display-item.citation.isi??? ND
social impact