{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,7]],"date-time":"2026-03-07T18:43:29Z","timestamp":1772909009274,"version":"3.50.1"},"publisher-location":"California","reference-count":0,"publisher":"International Joint Conferences on Artificial Intelligence Organization","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2017,8]]},"abstract":"<jats:p>Several recent works have empirically observed that Convolutional Neural Nets (CNNs) are (approximately) invertible. To understand this approximate invertibility phenomenon and how to leverage it more effectively, we focus on a theoretical explanation and develop a mathematical model of sparse signal recovery that is consistent with CNNs with random weights. We give an exact connection to a particular model of model-based compressive sensing (and its recovery algorithms) and random-weight CNNs. We show empirically that several learned networks are consistent with our mathematical analysis and then demonstrate that with such a simple theoretical framework, we can obtain reasonable reconstruction results on real images. We also discuss gaps between our model assumptions and the CNN trained for classification in practical scenarios.<\/jats:p>","DOI":"10.24963\/ijcai.2017\/236","type":"proceedings-article","created":{"date-parts":[[2017,7,28]],"date-time":"2017-07-28T09:14:07Z","timestamp":1501233247000},"page":"1703-1710","source":"Crossref","is-referenced-by-count":29,"title":["Towards Understanding the Invertibility of Convolutional Neural Networks"],"prefix":"10.24963","author":[{"given":"Anna","family":"Gilbert","sequence":"first","affiliation":[{"name":"University of Michigan"}]},{"given":"Yi","family":"Zhang","sequence":"additional","affiliation":[{"name":"University of Michigan"}]},{"given":"Kibok","family":"Lee","sequence":"additional","affiliation":[{"name":"University of Michigan"}]},{"given":"Yuting","family":"Zhang","sequence":"additional","affiliation":[{"name":"University of Michigan"}]},{"given":"Honglak","family":"Lee","sequence":"additional","affiliation":[{"name":"University of Michigan"},{"name":"Google"}]}],"member":"10584","event":{"name":"Twenty-Sixth International Joint Conference on Artificial Intelligence","theme":"Artificial Intelligence","location":"Melbourne, Australia","acronym":"IJCAI-2017","number":"26","sponsor":["International Joint Conferences on Artificial Intelligence Organization (IJCAI)","University of Technology Sydney (UTS)","Australian Computer Society (ACS)"],"start":{"date-parts":[[2017,8,19]]},"end":{"date-parts":[[2017,8,26]]}},"container-title":["Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence"],"original-title":[],"deposited":{"date-parts":[[2017,7,28]],"date-time":"2017-07-28T11:52:56Z","timestamp":1501242776000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ijcai.org\/proceedings\/2017\/236"}},"subtitle":[],"proceedings-subject":"Artificial Intelligence Research Articles","short-title":[],"issued":{"date-parts":[[2017,8]]},"references-count":0,"URL":"https:\/\/doi.org\/10.24963\/ijcai.2017\/236","relation":{},"subject":[],"published":{"date-parts":[[2017,8]]}}}