Beyond Graph Neural Networks with Lifted Relational Neural Networks

Explore a declarative differentiable programming framework based on Lifted Relational Neural Networks in this 15-minute conference talk. Delve into how small parameterized logic programs encode relational learning scenarios, covering topics such as symbolic vs. statistical AI, deep learning, relational representations, and statistical relational learning. Learn about declarative ANN encoding, multi-layer perceptrons, convolutional networks, and graph neural networks. Discover how this framework extends beyond traditional GNNs, offering insights into lossless model compression and increased expressiveness through atom rings. Access related resources, including the original paper, framework, and blog posts, to further expand your understanding of this cutting-edge approach in neuro-symbolic AI.

Intro
Symbolic vs. Statistical Al
Outline
Deep Learning
Relational Representations
Problem Statement
Statistical Relational Learning
Lifted Relational Neural Networks
Declarative ANN Encoding
Multi-Layer Perceptrons
Convolutional Networks
Recurrent and Recursive Networks
Graph Neural Networks
Lossless Model Compression via Lifting
Beyond GNN Expressiveness
Beyond GNNs with Atom Rings