·building·project·shipped

multi-graph-former

Like a transformer, but for graph2graph instead of seq2seq

  • Sequences are just a special class of graphs. The multi-graph-former can process any kind of graph.
  • Supports intra- and inter-graph attention, vert updates, and edge updates with dynamic structure
  • Implemented gated-update mechanism for both vertices and edges with einsum-based operations
  • Example applies multi-graph-former to building a graph-structured hidden state for a recurrent neural network that encodes a sequence of words

Neighborhood

Related

The Node Neural Network (NNN)The Node Neural Network...Broadening and Building Beyond Classical Reinforcement LearningBroadening and Building...Design Patterns for AIDesign Patterns for AIWhy I rebuilt the siteWhy I rebuilt the siteResearch and technical writingResearch and technical writ...multigraph-nnmultigraph-nnMulti-graph former — a transformer over heterogeneous graph structureMulti-graph former — a tran...multi-graph-former