Beyond graph neural networks with lifted relational neural networks

We introduce a declarative differentiable programming framework, based on the language of Lifted Relational Neural Networks, where small parameterized logic programs are used to encode deep relational learning scenarios through underlying symmetries. When presented with data, such as various forms graphs, program interpreter dynamically unfolds computation graphs be for parameter optimization by standard means. Following from declarative, logic-based encoding, this results into unified representation wide range neural models in form compact and elegant programs, contrast existing procedural approaches operating directly computational graph level. illustrate how idea can concise encoding advanced architectures, main focus Graph Networks (GNNs). Importantly, using we also show contemporary GNN easily extended towards higher expressiveness ways. In experiments, demonstrate correctness efficiency comparison against specialized frameworks, while shedding some light performance models.