Abstract Reasoning via Logic-guided Generation

Imitating humans in abstract reasoning is one of the aims of artificial intelligence. Previous research has used the  response elimination strategy (excluding candidate answers based on matching with the given context images) to train neural networks which solve problems resembling an IQ test. However, humans can also imagine the answer from context images without any candidates and select the most similar one.

Image credit: geralt via Pixabay (Free Pixabay licence)

Therefore, a recent paper on arXiv.org suggests reducing reasoning problems into optimization problems in propositional logic. Firstly, context images are embedded into propositional variables. Then, a differentiable reasoning layer predicts the variables of the answer image, and the decoder network generates the answer image.

It is shown that the framework performs comparably to neural networks that rely on response elimination despite not having access to the wrong candidates while training.

Abstract reasoning, i.e., inferring complicated patterns from given observations, is a central building block of artificial general intelligence. While humans find the answer by either eliminating wrong candidates or first constructing the answer, prior deep neural network (DNN)-based methods focus on the former discriminative approach. This paper aims to design a framework for the latter approach and bridge the gap between artificial and human intelligence. To this end, we propose logic-guided generation (LoGe), a novel generative DNN framework that reduces abstract reasoning as an optimization problem in propositional logic. LoGe is composed of three steps: extract propositional variables from images, reason the answer variables with a logic layer, and reconstruct the answer image from the variables. We demonstrate that LoGe outperforms the black box DNN frameworks for generative abstract reasoning under the RAVEN benchmark, i.e., reconstructing answers based on capturing correct rules of various attributes from observations.

Research paper: Yu, S., Mo, S., Ahn, S., and Shin, J., "Abstract Reasoning via Logic-guided Generation", 2021. Link: https://arxiv.org/abs/2107.10493