Solving Visual Analogies Using Neural Algorithmic Reasoning
This addresses a program synthesis task in visual reasoning, but it appears incremental as it builds on prior neural analogical reasoning methods.
The paper tackles the problem of visual analogical reasoning by discovering sequences of transformations between input/output image pairs to apply to new inputs, using a neural algorithmic reasoning approach that searches for neural network transformations on distributed representations. The result shows evaluation of generalization to images with unseen shapes and positions, but no concrete numbers are provided in the abstract.
We consider a class of visual analogical reasoning problems that involve discovering the sequence of transformations by which pairs of input/output images are related, so as to analogously transform future inputs. This program synthesis task can be easily solved via symbolic search. Using a variation of the `neural analogical reasoning' approach of (Velickovic and Blundell 2021), we instead search for a sequence of elementary neural network transformations that manipulate distributed representations derived from a symbolic space, to which input images are directly encoded. We evaluate the extent to which our `neural reasoning' approach generalizes for images with unseen shapes and positions.