Stephan Wäldchen
YOU?
Author Swipe
Hardness of Deceptive Certificate Selection Open
Recent progress towards theoretical interpretability guarantees for AI has been made with classifiers that are based on interactive proof systems. A prover selects a certificate from the datapoint and sends it to a verifier who decides the…
Interpretability Guarantees with Merlin-Arthur Classifiers Open
We propose an interactive multi-agent classifier that provides provable interpretability guarantees even for complex agents such as neural networks. These guarantees consist of lower bounds on the mutual information between selected featur…
View article: Training Characteristic Functions with Reinforcement Learning: XAI-methods play Connect Four
Training Characteristic Functions with Reinforcement Learning: XAI-methods play Connect Four Open
One of the goals of Explainable AI (XAI) is to determine which input components were relevant for a classifier decision. This is commonly know as saliency attribution. Characteristic functions (from cooperative game theory) are able to eva…
A Complete Characterisation of ReLU-Invariant Distributions Open
We give a complete characterisation of families of probability distributions that are invariant under the action of ReLU neural network layers. The need for such families arises during the training of Bayesian networks or the analysis of t…
A Rate-Distortion Framework for Explaining Neural Network Decisions Open
We formalise the widespread idea of interpreting neural network decisions as an explicit optimisation problem in a rate-distortion framework. A set of input features is deemed relevant for a classification decision if the expected classifi…
The Computational Complexity of Understanding Network Decisions Open
For a Boolean function $Φ\colon\{0,1\}^d\to\{0,1\}$ and an assignment to its variables $\mathbf{x}=(x_1, x_2, \dots, x_d)$ we consider the problem of finding the subsets of the variables that are sufficient to determine the function value …