XAI: Explainable AI for NN and Time Series

Neural networks are complex, non-linear models whose parameters are notoriously difficult to interpret — a limitation commonly referred to as the 'black box' problem. While Explainable AI (xAI) techniques have been developed to shed light on this opacity, classical xAI approaches are poorly suited to longitudinal data characterized by strict temporal dependencies. We propose a novel xAI framework specifically designed for neural networks operating in a time series context. Beyond enhancing interpretability, the framework enables real-time monitoring of the network's response to evolving market dynamics, thereby supporting risk-based applications — including adaptive on/off strategies contingent on the network's current operational state.

 

A dedicated tutorial is currently in preparation; corresponding references and links to the relevant literature will be made available in due course.