Figure 2 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
In this report we present a new model of concepts, based on the framework of variational autoencoders, which is designed to have attractive properties such as factored conceptual domains, and at the same time be learnable from data. The model is inspired by, and closely related to, the Beta-VAE model of concepts, but is designed to be more closely...
Contexts in source publication
Context 1
... flexibility could be useful when considering more realistic datasets, which may contain instances which are labelled with only a subset of the full set of atomic concepts. Figure 2 shows how to incorporate an additional label-the any label-into the model to allow this flexibility. The idea is that any missing labels will be replaced with the any label. ...
Context 2
... any, circle, any) which would denote a green circle of unspecified size and position. Since the atomic concepts are assumed to be uniformly distributed for a particular dimension, the graphical model in Figure 2 represents a Gaussian mixture model with equal probabilities assigned to each component of the mixture (so 1/3 in the case where there are three atomic labels for each domain, as we have below in our main dataset). ...
