LioNets (Local Interpretations Of Neural Networks through Penultimate Layer Decoding) LioNets is a methodology on providing explanations for a neural network's decisions, in a local scope, through a process that actively takes into consideration the neural network's architecture on creating an instance's neighbourhood, that assures the adjacency among the generated neighbours and the instance.
Additional information: In this asset, the code of LioNets is supporting a python notebook, which presents the technique applied to two different binary classification problems (Hate Speech and SMS Spam detection).
Updated version: LioNets V2.0 is also available on this platform. It provides better explanations, more examples on textual and time-series datasets.
The interpretation technique implemented in LioNets is intended to provide explanations to neural-based classifier, towards a trustworthy machine learning component. LioNets actively takes into consideration the internal structure of the model it tries to explain, providing that way better adjacency among the generated neighbours and the instance.
LioNets component is GDPR compliant (Articles 13–15) because it is providing a way to interpret the decision of a neural network in a form of feature importances. This approach addresses the "explicability" requirement of the GDPR, where a requirement is fixed for automated decision processes that have an impact on humans.