On the visual analytic intelligence of neural networks

TL;DR

This paper introduces a biologically realistic neural network system that processes synthetic eye movements to perform visual oddity tasks, surpassing human accuracy and demonstrating faster learning with fewer parameters.

Contribution

The paper presents a novel biologically inspired neural network architecture that incorporates eye movement dynamics and outperforms traditional models on visual oddity tasks.

Findings

Biologically inspired network achieves higher accuracy than conventional models.

The system learns faster and uses fewer parameters.

Both models share similar reasoning mechanisms.

Abstract

Visual oddity task was conceived as a universal ethnic-independent analytic intelligence test for humans. Advancements in artificial intelligence led to important breakthroughs, yet competing with humans on such analytic intelligence tasks remains challenging and typically resorts to non-biologically-plausible architectures. We present a biologically realistic system that receives inputs from synthetic eye movements - saccades, and processes them with neurons incorporating dynamics of neocortical neurons. We introduce a procedurally generated visual oddity dataset to train an architecture extending conventional relational networks and our proposed system. Both approaches surpass the human accuracy, and we uncover that both share the same essential underlying mechanism of reasoning. Finally, we show that the biologically inspired network achieves superior accuracy, learns faster and requires fewer parameters than the conventional network.