The term robustness is ubiquitous in modern Machine Learning (ML). However, its meaning varies depending on context and community. Researchers either focus on narrow technical definitions, such as adversarial robustness, natural distribution shifts, and performativity, or they simply leave open what exactly they mean by robustness. In this paper, we provide a conceptual analysis of the term robustness, with the aim to develop a common language, that allows us to weave together different strands of robustness research. We define robustness as (...) the relative stability of a robustness target with respect to specific interventions on a modifier. Our account captures the various sub-types of robustness that are discussed in the research literature, including robustness to distribution shifts, prediction robustness, or the robustness of algorithmic explanations. Finally, we delineate robustness from adjacent key concepts in ML, such as extrapolation, generalization, and uncertainty, and establish it as an independent epistemic concept. (shrink)

The same method that creates adversarial examples to fool image-classifiers can be used to generate counterfactual explanations that explain algorithmic decisions. This observation has led researchers to consider CEs as AEs by another name. We argue that the relationship to the true label and the tolerance with respect to proximity are two properties that formally distinguish CEs and AEs. Based on these arguments, we introduce CEs, AEs, and related concepts mathematically in a common framework. Furthermore, we show connections between current (...) methods for generating CEs and AEs, and estimate that the fields will merge more and more as the number of common use-cases grows. (shrink)

To learn about real world phenomena, scientists have traditionally used models with clearly interpretable elements. However, modern machine learning (ML) models, while powerful predictors, lack this direct elementwise interpretability (e.g. neural network weights). Interpretable machine learning (IML) offers a solution by analyzing models holistically to derive interpretations. Yet, current IML research is focused on auditing ML models rather than leveraging them for scientific inference. Our work bridges this gap, presenting a framework for designing IML methods—termed ’property descriptors’—that illuminate not just (...) the model, but also the phenomenon it represents. We demonstrate that property descriptors, grounded in statistical learning theory, can effectively reveal relevant properties of the joint probability distribution of the observational data. We identify existing IML methods suited for scientific inference and provide a guide for developing new descriptors with quantified epistemic uncertainty. Our framework empowers scientists to harness ML models for inference, and provides directions for future IML research to support scientific understanding. (shrink)