Matches in SemOpenAlex for { <https://semopenalex.org/work/W3034391323> ?p ?o ?g. }
- W3034391323 abstract "Injectivity plays an important role in generative models where it enables inference; in inverse problems and compressed sensing with generative priors it is a precursor to well posedness. We establish sharp characterizations of injectivity of fully-connected and convolutional ReLU layers and networks. First, through a layerwise analysis, we show that an expansivity factor of two is necessary and sufficient for injectivity by constructing appropriate weight matrices. We show that global injectivity with iid Gaussian matrices, a commonly used tractable model, requires larger expansivity between 3.4 and 10.5. We also characterize the stability of inverting an injective network via worst-case Lipschitz constants of the inverse. We then use arguments from differential topology to study injectivity of deep networks and prove that any Lipschitz map can be approximated by an injective ReLU network. Finally, using an argument based on random projections, we show that an end-to-end---rather than layerwise---doubling of the dimension suffices for injectivity. Our results establish a theoretical basis for the study of nonlinear inverse and inference problems using neural networks." @default.
- W3034391323 created "2020-06-19" @default.
- W3034391323 creator A5002015062 @default.
- W3034391323 creator A5014329642 @default.
- W3034391323 creator A5026803226 @default.
- W3034391323 creator A5052627284 @default.
- W3034391323 creator A5059920066 @default.
- W3034391323 date "2021-05-04" @default.
- W3034391323 modified "2023-09-27" @default.
- W3034391323 title "Globally Injective ReLU networks" @default.
- W3034391323 cites W143004564 @default.
- W3034391323 cites W1575147392 @default.
- W3034391323 cites W1583912456 @default.
- W3034391323 cites W1834627138 @default.
- W3034391323 cites W1965847424 @default.
- W3034391323 cites W1986736933 @default.
- W3034391323 cites W1988115241 @default.
- W3034391323 cites W2007626594 @default.
- W3034391323 cites W2025762001 @default.
- W3034391323 cites W2057678278 @default.
- W3034391323 cites W2093399550 @default.
- W3034391323 cites W2107179775 @default.
- W3034391323 cites W2122340891 @default.
- W3034391323 cites W2133740040 @default.
- W3034391323 cites W2158581396 @default.
- W3034391323 cites W2173520492 @default.
- W3034391323 cites W2194775991 @default.
- W3034391323 cites W2318895797 @default.
- W3034391323 cites W2409550820 @default.
- W3034391323 cites W2411541852 @default.
- W3034391323 cites W2412320034 @default.
- W3034391323 cites W2502312327 @default.
- W3034391323 cites W2524985544 @default.
- W3034391323 cites W2528305538 @default.
- W3034391323 cites W2595294663 @default.
- W3034391323 cites W2617427826 @default.
- W3034391323 cites W2752851182 @default.
- W3034391323 cites W2785678896 @default.
- W3034391323 cites W2787898208 @default.
- W3034391323 cites W2788549480 @default.
- W3034391323 cites W2795783309 @default.
- W3034391323 cites W2810203579 @default.
- W3034391323 cites W2883105305 @default.
- W3034391323 cites W2895434480 @default.
- W3034391323 cites W2945125725 @default.
- W3034391323 cites W2949117887 @default.
- W3034391323 cites W2949366180 @default.
- W3034391323 cites W2951085270 @default.
- W3034391323 cites W2952020389 @default.
- W3034391323 cites W2952568974 @default.
- W3034391323 cites W2954241485 @default.
- W3034391323 cites W2962770929 @default.
- W3034391323 cites W2962879692 @default.
- W3034391323 cites W2963030988 @default.
- W3034391323 cites W2963139417 @default.
- W3034391323 cites W2963226019 @default.
- W3034391323 cites W2963300719 @default.
- W3034391323 cites W2963341557 @default.
- W3034391323 cites W2963373786 @default.
- W3034391323 cites W2963684275 @default.
- W3034391323 cites W2963685250 @default.
- W3034391323 cites W2963919714 @default.
- W3034391323 cites W2963981733 @default.
- W3034391323 cites W2964294232 @default.
- W3034391323 cites W2970188304 @default.
- W3034391323 cites W3007259350 @default.
- W3034391323 cites W3007572682 @default.
- W3034391323 cites W3033846281 @default.
- W3034391323 cites W3035626604 @default.
- W3034391323 cites W3037212379 @default.
- W3034391323 cites W3037604831 @default.
- W3034391323 cites W3125537303 @default.
- W3034391323 hasPublicationYear "2021" @default.
- W3034391323 type Work @default.
- W3034391323 sameAs 3034391323 @default.
- W3034391323 citedByCount "7" @default.
- W3034391323 countsByYear W30343913232021 @default.
- W3034391323 crossrefType "journal-article" @default.
- W3034391323 hasAuthorship W3034391323A5002015062 @default.
- W3034391323 hasAuthorship W3034391323A5014329642 @default.
- W3034391323 hasAuthorship W3034391323A5026803226 @default.
- W3034391323 hasAuthorship W3034391323A5052627284 @default.
- W3034391323 hasAuthorship W3034391323A5059920066 @default.
- W3034391323 hasConcept C105795698 @default.
- W3034391323 hasConcept C107673813 @default.
- W3034391323 hasConcept C11413529 @default.
- W3034391323 hasConcept C114614502 @default.
- W3034391323 hasConcept C121332964 @default.
- W3034391323 hasConcept C128107574 @default.
- W3034391323 hasConcept C154945302 @default.
- W3034391323 hasConcept C163716315 @default.
- W3034391323 hasConcept C177769412 @default.
- W3034391323 hasConcept C184720557 @default.
- W3034391323 hasConcept C202444582 @default.
- W3034391323 hasConcept C207467116 @default.
- W3034391323 hasConcept C22324862 @default.
- W3034391323 hasConcept C2524010 @default.
- W3034391323 hasConcept C2776214188 @default.
- W3034391323 hasConcept C28826006 @default.
- W3034391323 hasConcept C33676613 @default.