GNN for Supply Chain Modeling?

Hello TF community!

I’m looking to model a company’s supply chain in order to optimize distribution of products from their warehouses to their retail outlets. The company currently relies on linear programming methods to make these optimizations, and they’re interested in ML approaches which could scale better for their sizable volume.

Considering a distribution network comprises of nodes (type a: warehouse; type b: retail outlet), and flows between these nodes (unidirectional flow of products from warehouse to retailer), this seems like a ripe opportunity to use graph neural networks, such as TF’s newly released TF-GNN package.

Therefore, if the objective is to maximize for something like % of retail outlet space occupied at any given time (aiming for >90% stock of products), how can I model this problem using TF-GNN?

Would you choose a different approach entirely? Curious for more perspective.

There was a nice survey about GNNs for Combinatorial Optimization at:

https://arxiv.org/abs/2102.09544

Take a look also at: