Making sense of CNNs: interpreting deep representations & their invariances with INNs

To tackle increasingly complex tasks, it has become an essential ability of neural networks to learn abstract representations. These task-specific representations and, particularly, the invariances they capture turn neural networks into black box models that lack interpretability. To open such a bla...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Rombach, Robin (VerfasserIn) , Esser, Patrick (VerfasserIn) , Ommer, Björn (VerfasserIn)
Dokumenttyp: Article (Journal) Kapitel/Artikel
Sprache:Englisch
Veröffentlicht: 4 Aug 2020
In: Arxiv
Year: 2020, Pages: 1-41
DOI:10.48550/arXiv.2008.01777
Online-Zugang:Verlag, lizenzpflichtig, Volltext: https://doi.org/10.48550/arXiv.2008.01777
Verlag, lizenzpflichtig, Volltext: http://arxiv.org/abs/2008.01777
Volltext
Verfasserangaben:Robin Rombach, Patrick Esser, and Björn Ommer

MARC

LEADER 00000caa a2200000 c 4500
001 1818105357
003 DE-627
005 20230118164611.0
007 cr uuu---uuuuu
008 221006s2020 xx |||||o 00| ||eng c
024 7 |a 10.48550/arXiv.2008.01777  |2 doi 
035 |a (DE-627)1818105357 
035 |a (DE-599)KXP1818105357 
035 |a (OCoLC)1361714928 
040 |a DE-627  |b ger  |c DE-627  |e rda 
041 |a eng 
084 |a 27  |2 sdnb 
100 1 |a Rombach, Robin  |e VerfasserIn  |0 (DE-588)1269607952  |0 (DE-627)1818109115  |4 aut 
245 1 0 |a Making sense of CNNs  |b interpreting deep representations & their invariances with INNs  |c Robin Rombach, Patrick Esser, and Björn Ommer 
246 3 0 |a and 
264 1 |c 4 Aug 2020 
300 |a 41 
336 |a Text  |b txt  |2 rdacontent 
337 |a Computermedien  |b c  |2 rdamedia 
338 |a Online-Ressource  |b cr  |2 rdacarrier 
500 |a Gesehen am 06.10.2022 
520 |a To tackle increasingly complex tasks, it has become an essential ability of neural networks to learn abstract representations. These task-specific representations and, particularly, the invariances they capture turn neural networks into black box models that lack interpretability. To open such a black box, it is, therefore, crucial to uncover the different semantic concepts a model has learned as well as those that it has learned to be invariant to. We present an approach based on INNs that (i) recovers the task-specific, learned invariances by disentangling the remaining factor of variation in the data and that (ii) invertibly transforms these recovered invariances combined with the model representation into an equally expressive one with accessible semantic concepts. As a consequence, neural network representations become understandable by providing the means to (i) expose their semantic meaning, (ii) semantically modify a representation, and (iii) visualize individual learned semantic concepts and invariances. Our invertible approach significantly extends the abilities to understand black box models by enabling post-hoc interpretations of state-of-the-art networks without compromising their performance. Our implementation is available at https://compvis.github.io/invariances/ . 
650 4 |a Computer Science - Computer Vision and Pattern Recognition 
700 1 |a Esser, Patrick  |e VerfasserIn  |0 (DE-588)1116611430  |0 (DE-627)870571214  |0 (DE-576)478524323  |4 aut 
700 1 |a Ommer, Björn  |d 1981-  |e VerfasserIn  |0 (DE-588)1034893106  |0 (DE-627)746457510  |0 (DE-576)382507916  |4 aut 
773 0 8 |i Enthalten in  |t Arxiv  |d Ithaca, NY : Cornell University, 1991  |g (2020), Artikel-ID 2008.01777, Seite 1-41  |h Online-Ressource  |w (DE-627)509006531  |w (DE-600)2225896-6  |w (DE-576)28130436X  |7 nnas  |a Making sense of CNNs interpreting deep representations & their invariances with INNs 
773 1 8 |g year:2020  |g elocationid:2008.01777  |g pages:1-41  |g extent:41  |a Making sense of CNNs interpreting deep representations & their invariances with INNs 
856 4 0 |u https://doi.org/10.48550/arXiv.2008.01777  |x Verlag  |x Resolving-System  |z lizenzpflichtig  |3 Volltext 
856 4 0 |u http://arxiv.org/abs/2008.01777  |x Verlag  |z lizenzpflichtig  |3 Volltext 
951 |a AR 
992 |a 20221006 
993 |a Article 
994 |a 2020 
998 |g 1034893106  |a Ommer, Björn  |m 1034893106:Ommer, Björn  |d 700000  |d 708070  |d 700000  |d 728500  |e 700000PO1034893106  |e 708070PO1034893106  |e 700000PO1034893106  |e 728500PO1034893106  |k 0/700000/  |k 1/700000/708070/  |k 0/700000/  |k 1/700000/728500/  |p 3  |y j 
998 |g 1116611430  |a Esser, Patrick  |m 1116611430:Esser, Patrick  |d 700000  |d 708070  |e 700000PE1116611430  |e 708070PE1116611430  |k 0/700000/  |k 1/700000/708070/  |p 2 
998 |g 1269607952  |a Rombach, Robin  |m 1269607952:Rombach, Robin  |d 700000  |d 708070  |d 700000  |d 728500  |e 700000PR1269607952  |e 708070PR1269607952  |e 700000PR1269607952  |e 728500PR1269607952  |k 0/700000/  |k 1/700000/708070/  |k 0/700000/  |k 1/700000/728500/  |p 1  |x j 
999 |a KXP-PPN1818105357  |e 4194706013 
BIB |a Y 
JSO |a {"physDesc":[{"extent":"41 S."}],"relHost":[{"id":{"eki":["509006531"],"zdb":["2225896-6"]},"origin":[{"dateIssuedDisp":"1991-","publisher":"Cornell University ; Arxiv.org","dateIssuedKey":"1991","publisherPlace":"Ithaca, NY ; [Erscheinungsort nicht ermittelbar]"}],"physDesc":[{"extent":"Online-Ressource"}],"title":[{"title_sort":"Arxiv","title":"Arxiv"}],"language":["eng"],"recId":"509006531","note":["Gesehen am 28.05.2024"],"disp":"Making sense of CNNs interpreting deep representations & their invariances with INNsArxiv","type":{"media":"Online-Ressource","bibl":"edited-book"},"part":{"extent":"41","text":"(2020), Artikel-ID 2008.01777, Seite 1-41","pages":"1-41","year":"2020"},"titleAlt":[{"title":"Arxiv.org"},{"title":"Arxiv.org e-print archive"},{"title":"Arxiv e-print archive"},{"title":"De.arxiv.org"}],"pubHistory":["1991 -"]}],"origin":[{"dateIssuedKey":"2020","dateIssuedDisp":"4 Aug 2020"}],"id":{"eki":["1818105357"],"doi":["10.48550/arXiv.2008.01777"]},"name":{"displayForm":["Robin Rombach, Patrick Esser, and Björn Ommer"]},"note":["Gesehen am 06.10.2022"],"type":{"media":"Online-Ressource","bibl":"chapter"},"language":["eng"],"recId":"1818105357","title":[{"subtitle":"interpreting deep representations & their invariances with INNs","title":"Making sense of CNNs","title_sort":"Making sense of CNNs"}],"person":[{"roleDisplay":"VerfasserIn","display":"Rombach, Robin","role":"aut","family":"Rombach","given":"Robin"},{"display":"Esser, Patrick","roleDisplay":"VerfasserIn","role":"aut","family":"Esser","given":"Patrick"},{"given":"Björn","family":"Ommer","role":"aut","display":"Ommer, Björn","roleDisplay":"VerfasserIn"}]} 
SRT |a ROMBACHROBMAKINGSENS4202