[PDF] A simple neural network module for relational reasoning | Semantic Scholar

R recurrent relational networks are introduced which increase the suite of solvable tasks to those that require an order of magnitude more steps of relational reasoning and are applied to the BaBi textual QA dataset solving 19/20 tasks.

The recurrent relational network is introduced, a general purpose module that operates on a graph representation of objects that can augment any neural network model with the capacity to do many-step relational reasoning.

Stacked Attention Recurrent Relational Networks (SARRN) is introduced to answer natural language questions from facts, which fundamentally hinge on multiple steps of relational reasoning and improve the ability of reasoning.

The Working Memory Network is introduced, a MemNN architecture with a novel working memory storage and reasoning module that retains the relational reasoning abilities of the RN while reducing its computational complexity from quadratic to linear.

This paper describes an architecture that enables relationships to be determined from a stream of entities obtained by an attention mechanism over the input field, and demonstrates equivalent performance with greater interpretability while requiring only a fraction of the model parameters of the original RN module.

A new memory module — a \textit{Relational Memory Core} (RMC) — is used which employs multi-head dot product attention to allow memories to interact and achieves state-of-the-art results on the WikiText-103, Project Gutenberg, and GigaWord datasets.

This work proposes a simple and general network module called a Set Refiner Network (SRN) and inserts the module into existing relational reasoning models and shows that respecting set invariance leads to substantial gains in prediction performance and robustness on several relational reasoning tasks.

A novel neural modular approach that performs compositional reasoning by automatically inducing a desired sub-task decomposition without relying on strong supervision is presented, which is more interpretable to human evaluators compared to other state-of-the-art models.

Finding ReMO (Related Memory Object): A Simple Neural Architecture for Text based Reasoning

Memory Network based models have shown a remarkable progress on the task of relational reasoning. Recently, a simpler yet powerful neural network module called Relation Network (RN) has been

…