Matches in SemOpenAlex for { <https://semopenalex.org/work/W2896687685> ?p ?o ?g. }
- W2896687685 abstract "A neural network deployed in the wild may be asked to make predictions for inputs that were drawn from a different distribution than that of the training data. A plethora of work has demonstrated that it is easy to find or synthesize inputs for which a neural network is highly confident yet wrong. Generative models are widely viewed to be robust to such mistaken confidence as modeling the density of the input features can be used to detect novel, out-of-distribution inputs. In this paper we challenge this assumption. We find that the density learned by flow-based models, VAEs, and PixelCNNs cannot distinguish images of common objects such as dogs, trucks, and horses (i.e. CIFAR-10) from those of house numbers (i.e. SVHN), assigning a higher likelihood to the latter when the model is trained on the former. Moreover, we find evidence of this phenomenon when pairing several popular image data sets: FashionMNIST vs MNIST, CelebA vs SVHN, ImageNet vs CIFAR-10 / CIFAR-100 / SVHN. To investigate this curious behavior, we focus analysis on flow-based generative models in particular since they are trained and evaluated via the exact marginal likelihood. We find such behavior persists even when we restrict the flows to constant-volume transformations. These transformations admit some theoretical analysis, and we show that the difference in likelihoods can be explained by the location and variances of the data and the model curvature. Our results caution against using the density estimates from deep generative models to identify inputs similar to the training distribution until their behavior for out-of-distribution inputs is better understood." @default.
- W2896687685 created "2018-10-26" @default.
- W2896687685 creator A5003054496 @default.
- W2896687685 creator A5008645497 @default.
- W2896687685 creator A5012859570 @default.
- W2896687685 creator A5064373793 @default.
- W2896687685 creator A5081258844 @default.
- W2896687685 date "2018-10-22" @default.
- W2896687685 modified "2023-10-16" @default.
- W2896687685 title "Do Deep Generative Models Know What They Don't Know?" @default.
- W2896687685 cites W1959608418 @default.
- W2896687685 cites W1981025032 @default.
- W2896687685 cites W2099471712 @default.
- W2896687685 cites W2104067967 @default.
- W2896687685 cites W2111406701 @default.
- W2896687685 cites W2123098109 @default.
- W2896687685 cites W2126022166 @default.
- W2896687685 cites W2149950545 @default.
- W2896687685 cites W2335728318 @default.
- W2896687685 cites W2531327146 @default.
- W2896687685 cites W2537458122 @default.
- W2896687685 cites W2559944883 @default.
- W2896687685 cites W2592505114 @default.
- W2896687685 cites W2594717275 @default.
- W2896687685 cites W2788017640 @default.
- W2896687685 cites W2883243526 @default.
- W2896687685 cites W2892375480 @default.
- W2896687685 cites W2895363208 @default.
- W2896687685 cites W2904981516 @default.
- W2896687685 cites W2949382160 @default.
- W2896687685 cites W2962695743 @default.
- W2896687685 cites W2962897886 @default.
- W2896687685 cites W2963139417 @default.
- W2896687685 cites W2963238274 @default.
- W2896687685 cites W2963384319 @default.
- W2896687685 cites W2963693742 @default.
- W2896687685 cites W2963857374 @default.
- W2896687685 cites W2964020555 @default.
- W2896687685 cites W2964153729 @default.
- W2896687685 cites W3118608800 @default.
- W2896687685 hasPublicationYear "2018" @default.
- W2896687685 type Work @default.
- W2896687685 sameAs 2896687685 @default.
- W2896687685 citedByCount "26" @default.
- W2896687685 countsByYear W28966876852019 @default.
- W2896687685 countsByYear W28966876852020 @default.
- W2896687685 countsByYear W28966876852021 @default.
- W2896687685 crossrefType "posted-content" @default.
- W2896687685 hasAuthorship W2896687685A5003054496 @default.
- W2896687685 hasAuthorship W2896687685A5008645497 @default.
- W2896687685 hasAuthorship W2896687685A5012859570 @default.
- W2896687685 hasAuthorship W2896687685A5064373793 @default.
- W2896687685 hasAuthorship W2896687685A5081258844 @default.
- W2896687685 hasConcept C115961682 @default.
- W2896687685 hasConcept C119857082 @default.
- W2896687685 hasConcept C154945302 @default.
- W2896687685 hasConcept C167966045 @default.
- W2896687685 hasConcept C190502265 @default.
- W2896687685 hasConcept C2524010 @default.
- W2896687685 hasConcept C2984842247 @default.
- W2896687685 hasConcept C33923547 @default.
- W2896687685 hasConcept C38349280 @default.
- W2896687685 hasConcept C39890363 @default.
- W2896687685 hasConcept C41008148 @default.
- W2896687685 hasConcept C50644808 @default.
- W2896687685 hasConceptScore W2896687685C115961682 @default.
- W2896687685 hasConceptScore W2896687685C119857082 @default.
- W2896687685 hasConceptScore W2896687685C154945302 @default.
- W2896687685 hasConceptScore W2896687685C167966045 @default.
- W2896687685 hasConceptScore W2896687685C190502265 @default.
- W2896687685 hasConceptScore W2896687685C2524010 @default.
- W2896687685 hasConceptScore W2896687685C2984842247 @default.
- W2896687685 hasConceptScore W2896687685C33923547 @default.
- W2896687685 hasConceptScore W2896687685C38349280 @default.
- W2896687685 hasConceptScore W2896687685C39890363 @default.
- W2896687685 hasConceptScore W2896687685C41008148 @default.
- W2896687685 hasConceptScore W2896687685C50644808 @default.
- W2896687685 hasLocation W28966876851 @default.
- W2896687685 hasOpenAccess W2896687685 @default.
- W2896687685 hasPrimaryLocation W28966876851 @default.
- W2896687685 hasRelatedWork W1959608418 @default.
- W2896687685 hasRelatedWork W2099471712 @default.
- W2896687685 hasRelatedWork W2108598243 @default.
- W2896687685 hasRelatedWork W2117539524 @default.
- W2896687685 hasRelatedWork W2194775991 @default.
- W2896687685 hasRelatedWork W2335728318 @default.
- W2896687685 hasRelatedWork W2409550820 @default.
- W2896687685 hasRelatedWork W2531327146 @default.
- W2896687685 hasRelatedWork W2750384547 @default.
- W2896687685 hasRelatedWork W2867167548 @default.
- W2896687685 hasRelatedWork W2904981516 @default.
- W2896687685 hasRelatedWork W2913300775 @default.
- W2896687685 hasRelatedWork W2963238274 @default.
- W2896687685 hasRelatedWork W2963546708 @default.
- W2896687685 hasRelatedWork W2963693742 @default.
- W2896687685 hasRelatedWork W2964059111 @default.
- W2896687685 hasRelatedWork W2964212410 @default.
- W2896687685 hasRelatedWork W2970946347 @default.
- W2896687685 hasRelatedWork W2980402966 @default.
- W2896687685 hasRelatedWork W3118608800 @default.