🤖 AI Summary
The terms “ensemble” and “ensembling” suffer from semantic ambiguity, heterogeneous objectives, and fragmented practices across machine learning, statistics, and climate modeling.
Method: We propose the first interdisciplinary taxonomy of ensemble concepts, integrating conceptual analysis, terminological modeling, and cross-domain case comparison to construct a context-sensitive, semantically decoupled typology—eschewing prescriptive terminological unification in favor of consensus-driven collaboration.
Contribution/Results: Our framework systematically uncovers fundamental distinctions across four dimensions: (i) purpose (e.g., uncertainty quantification vs. performance enhancement), (ii) constituent units (models, data, or parameters), (iii) combination logic (e.g., weighted averaging, voting, or dynamical coupling), and (iv) evaluation paradigms. It enhances researchers’ conceptual clarity and cross-disciplinary communication efficiency, while providing a scalable theoretical foundation and practical guidance for method transfer and collaborative innovation.
📝 Abstract
When we think of model ensembling or ensemble modeling, there are many possibilities that come to mind in different disciplines. For example, one might think of a set of descriptions of a phenomenon in the world, perhaps a time series or a snapshot of multivariate space, and perhaps that set is comprised of data-independent descriptions, or perhaps it is quite intentionally fit *to* data, or even a suite of data sets with a common theme or intention. The very meaning of 'ensemble' - a collection together - conjures different ideas across and even within disciplines approaching phenomena. In this paper, we present a typology of the scope of these potential perspectives. It is not our goal to present a review of terms and concepts, nor is it to convince all disciplines to adopt a common suite of terms, which we view as futile. Rather, our goal is to disambiguate terms, concepts, and processes associated with 'ensembles' and 'ensembling' in order to facilitate communication, awareness, and possible adoption of tools across disciplines.