Compositional Attention Networks for Machine Reasoning

TLDR

MAC moves away from monolithic black‑box neural architectures toward a design that encourages both transparency and versatility. We present the MAC network, a novel fully differentiable neural network architecture designed to facilitate explicit and expressive reasoning. The model decomposes problems into a series of attention‑based reasoning steps executed by recurrent MAC cells that separate control and memory, and by stringing these cells together with structural constraints, it learns iterative reasoning processes directly from data in an end‑to‑end fashion. On the CLEVR visual reasoning dataset, the MAC network achieves 98.9 % accuracy—halving the error rate of the previous best model—while also being computationally and data efficient, requiring five times less data to reach strong results.

Abstract

We present the MAC network, a novel fully differentiable neural network architecture, designed to facilitate explicit and expressive reasoning. MAC moves away from monolithic black-box neural architectures towards a design that encourages both transparency and versatility. The model approaches problems by decomposing them into a series of attention-based reasoning steps, each performed by a novel recurrent Memory, Attention, and Composition (MAC) cell that maintains a separation between control and memory. By stringing the cells together and imposing structural constraints that regulate their interaction, MAC effectively learns to perform iterative reasoning processes that are directly inferred from the data in an end-to-end approach. We demonstrate the model's strength, robustness and interpretability on the challenging CLEVR dataset for visual reasoning, achieving a new state-of-the-art 98.9% accuracy, halving the error rate of the previous best model. More importantly, we show that the model is computationally-efficient and data-efficient, in particular requiring 5x less data than existing models to achieve strong results.