B C D(1) Accuracies
Source ModelTarget Model(2) Class-level Performance (2.1) High accuracies on both models (2.2) Big difference between the source and the target model accuracies (2.3) t-SNE Projection Result (3.1) Rankings of Important Neurons (3) Network Relation View (3.3) Weight Visualization for Layer 7 (Less weights in the target model have correspondences in the source side) (4) Domain Discriminability (4.1) Neurons (4.2) Images (3.2) Many Target Weights with Source Correspondence (4.3) Domain-invariant Feature C.1 C.2 Fig. 1. The visual analytics for transfer learning interface consists of four components: (A) statistical information summary, (B) the instance view, (C) the network relation view, and (D) the feature view. For the Office-31 dataset, (1) the prediction accuracy of the target model on the source dataset is lower than the source model; (2) for the target dataset, classes such as file cabinet and phone have a large performance gap between the source and the target models; (3) the neuron similarity matrices and the weights are presented; (4) some neurons have high domain discriminability, while Neuron #173 in Layer 5 is domain-invariant.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.