Ancona
On the Robustness of Removal-Based Feature Attributions Chris Lin
In recent years, machine learning has shown great promise in a variety of real-world applications.
Granger Components Analysis: Unsupervised learning of latent temporal dependencies
Here the concept of Granger causality is employed to propose a new criterion for unsupervised learning that is appropriate in the case of temporally-dependent source signals. The basic idea is to identify two projections of a multivariate time series such that the Granger causality among the resulting pair of components is maximized.
Benchmarking the Attribution Quality of Vision Models Robin Hesse 1 Simone Schaub-Meyer 1,2 Stefan Roth 1,2 1 Department of Computer Science, Technical University of Darmstadt
Attribution maps are one of the most established tools to explain the functioning of computer vision models. They assign importance scores to input features, indicating how relevant each feature is for the prediction of a deep neural network.
A Holistic Approach to Unifying Automatic Concept Extraction and Concept Importance Estimation
In recent years, concept-based approaches have emerged as some of the most promising explainability methods to help us interpret the decisions of Artificial Neural Networks (ANNs).
Understanding Visual Feature Reliance through the Lens of Complexity Thomas Fel Google DeepMind Brown University Louis Béthune
Recent studies suggest that deep learning models' inductive bias towards favoring
Towards the Difficulty for a Deep Neural Network to Learn Concepts of Different Complexities
DNN to represent the collaboration between a set of input variables.