Introduced in Modeling Uncertainty with Hedged Instance Embedding
N-Digit MNIST is a multi-digit MNIST-like dataset.
The hedged instance embedding (HIB) is introduced in which embeddings are modeled as random variables and the model is trained under the variational information bottleneck principle and results in improved performance for image matching and classification tasks, more structure in the learned embedding space, and an ability to compute a per-exemplar uncertainty measure that is correlated with downstream performance.
