DeepMind Introduces Relational RNN: A Tool for Reinforcement Learning
New Intelligence Compilation Source: arxiv Editor: Xiao Qin 【New Intelligence Guide】Traditional memory architectures struggle with relational reasoning. This paper from DeepMind and University College London proposes the Relational Memory Core (RMC), which can perform relational reasoning within sequential information, achieving state-of-the-art performance on the WikiText-103, Project Gutenberg, and GigaWord datasets. Paper: https://arxiv.org/pdf/1806.01822v1.pdf Memory-based neural networks … Read more
