Loss convergence in a causal Bayesian neural network of retail firm performance

TL;DR

This paper investigates the convergence behavior of causal Bayesian neural networks in retail firm performance modeling, demonstrating that removing weak nodes and using specific inference methods can enhance neural network convergence.

Contribution

It introduces a causal Bayesian neural network approach for retail performance analysis and explores how different inference techniques affect convergence, extending prior SEM results.

Findings

Removing weak SEM nodes improves convergence.

Flipout layers enhance neural network stability.

Vadam optimizer results are inconclusive.

Abstract

We extend the empirical results from the structural equation model (SEM) published in the paper Assortment Planning for Retail Buying, Retail Store Operations, and Firm Performance [1] by implementing the directed acyclic graph as a causal Bayesian neural network. Neural network convergence is shown to improve with the removal of the node with the weakest SEM path when variational inference is provided by perturbing weights with Flipout layers, while results from perturbing weights at the output with the Vadam optimizer are inconclusive.