ANC Workshop: Charlie Nash, Chair: Heru Praptono

Title:

The Multi-Entity Variational Autoencoder

Abstract:

Representing the world as objects is core to human intelligence. It is the basis of people’s sophisticated capacities for reasoning, imagining, planning, and learning. Artificial intelligence typically assumes human-defined representations of objects, and little work has explored how object-based representations can arise through unsupervised learning. Here we present an approach for learning probabilistic, object-based representations from data, called the “multi-entity variational autoencoder” (MVAE), whose prior and posterior distributions are defined over a set of random vectors. We demonstrate that the model can learn interpretable representations of visual scenes that disentangle objects and their properties.