Object-Centric Learning as Nested Optimization

Michael Chang, Sergey Levine, Thomas L. Griffiths

Keywords: nested optimization, iterative amortized inference, object-centric learning

TL;DR: Our primary contribution is a proposal for a unifying tangible problem statement under which to view the set of iterative approaches to developed for object-centric learning so far.

Abstract:

Various iterative algorithms have shown promising results in unsupervised decomposition simple visual scenes into representations of humans could intuitively consider objects, but all with different algorithmic and implementational design choices for making them work.
In this paper, we ask what the underlying computational problem that all of these iterative approaches are solving.
We show that these approaches can all be viewed as instances of algorithms for solving a particular nested optimization problem whose inner optimization is that of maximizing the ELBO with respect to a set of independently initialized parameters for each datapoint.
We lastly discuss how our nested optimization formulation reveals connections to similar problems studied in other fields, enabling us to leverage tools developed in these other fields to improve our object-centric learning methods.