Fabian Fumagalli

Adaptive coalition selection in ShaplEIG boosts Shapley value estimation efficiency, slashing computational complexity and enhancing performance in resource-constrained settings.

ProxySHAP slashes the computational cost of Shapley interaction estimation while simultaneously boosting accuracy, finally making high-order interaction analysis practical for models with thousands of features.

Additive explanations of survival models fail because of their inherent non-additivity, but now there's a way to decompose feature interactions into time-dependent components to understand *when* and *why* they fail.