Event info

The wealth of features: towards a coherent cooperative game theory for feature importance

Date: Tuesday, 7 May 2024 12:00 - 13:00
Place: Online via Zoom
Speaker: Daniel Fryer (La Trobe University)

Abstract:
Over the last decade, the integration of cooperative game theory, particularly Shapley values,into feature importance for explainable AI has gained traction. This approach offers axiomatic guarantees, suggesting a principled distribution of importance among features. However, the focus has largely remained on Shapley values and transferable utility games, limiting exploration into alternative game structures and solution concepts. This talk explores a broader scope, including different game classes, axioms and characteristic functions. Through various angles we challenge the standard reliance on Shapley values, advocating for a broader, more nuanced application of game theory.