RELAGRAM: Recursive Latent Graph Model for the Capacitated Vehicle Routing Problem

Key points

• Latent recursion repeatedly processes internal representations through the same neural network.1
• Recent work has shown this approach to be effective on structured reasoning tasks such as Sudoku and maze solving.1
• However, whether latent recursion can support neural routing remains unclear.1
• RELAGRAM encodes a capacitated vehicle routing problem (CVRP) instance as edge tokens; a compact transformer recurrently updates the latent representations of these tokens.1

Sources (1)

• [1]RELAGRAM: Recursive Latent Graph Model for the Capacitated Vehicle Routing Problem
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 10, 06:29 PM

  We introduce RELAGRAM (Recursive Latent Graph Model), a novel approach that brings latent recursion to graph combinatorial optimization.

  Latent recursion repeatedly processes internal representations through the same neural network.

Extractive summary: sentences quoted from the sources.