Journal Article
Knowledge Graphs

Informed Machine Learning - A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems

Laura von Rueden(Fraunhofer Institute for Intelligent Analysis and Information Systems), Sebastian Mayer(Fraunhofer Institute for Algorithms and Scientific Computing), Katharina Beckh(Fraunhofer Institute for Intelligent Analysis and Information Systems), Bogdan Georgiev(Fraunhofer Institute for Intelligent Analysis and Information Systems), Sven Giesselbach(Fraunhofer Institute for Intelligent Analysis and Information Systems), Raoul Heese(Fraunhofer Institute for Industrial Mathematics), Birgit Kirsch(Fraunhofer Institute for Intelligent Analysis and Information Systems), Michał Walczak(Fraunhofer Institute for Industrial Mathematics), Julius Pfrommer(Fraunhofer Institute of Optronics, System Technologies and Image Exploitation), Annika Pick(Fraunhofer Institute for Intelligent Analysis and Information Systems), Rajkumar Ramamurthy(Fraunhofer Institute for Intelligent Analysis and Information Systems), Jochen Garcke(Fraunhofer Institute for Algorithms and Scientific Computing), Christian Bauckhage(Fraunhofer Institute for Intelligent Analysis and Information Systems), Jannis Schuecker(Fraunhofer Institute for Machine Tools and Forming Technology)
January 1, 2021IEEE Transactions on Knowledge and Data Engineering877 citations
arXivPDFDOI

877

Citations

39

Influential Citations

IEEE Transactions on Knowledge and Data Engineering

Venue

2021

Year

Abstract

Despite its great success, machine learning can have its limits when dealing with insufficient training data. A potential solution is the additional integration of prior knowledge into the training process which leads to the notion of informed machine learning. In this paper, we present a structured overview of various approaches in this field. We provide a definition and propose a concept for informed machine learning which illustrates its building blocks and distinguishes it from conventional machine learning. We introduce a taxonomy that serves as a classification framework for informed machine learning approaches. It considers the source of knowledge, its representation, and its integration into the machine learning pipeline. Based on this taxonomy, we survey related research and describe how different knowledge representations such as algebraic equations, logic rules, or simulation results can be used in learning systems. This evaluation of numerous papers on the basis of our taxonomy uncovers key methods in the field of informed machine learning.

Analysis

Why This Paper Matters

This paper addresses a critical bottleneck in modern machine learning: the reliance on large, labeled datasets. In many real-world applications—such as healthcare, physics, or engineering—data is scarce, expensive, or noisy. The concept of informed machine learning offers a principled way to compensate by injecting prior knowledge (e.g., physical laws, expert rules, or simulation outputs) directly into the learning process. By providing a clear taxonomy and survey, the authors give practitioners a roadmap for choosing and combining knowledge integration strategies.

The paper is especially relevant for knowledge graph practitioners, as it explicitly covers logic rules and relational knowledge as forms of prior knowledge. It bridges symbolic AI and statistical learning, a key trend in neuro-symbolic systems. The taxonomy helps demystify the landscape, making it easier to compare approaches like physics-informed neural networks, rule-based regularization, or knowledge graph embeddings.

Technical Contributions

  • Definition and conceptual model: Clearly defines informed ML and its building blocks (knowledge source, representation, integration point).
  • Taxonomy: Three dimensions—knowledge source (data, human, simulation), representation (algebraic, logic, probabilistic, etc.), and integration (training data, loss function, architecture, etc.).
  • Comprehensive survey: Classifies dozens of papers, highlighting representative methods for each combination.
  • Identification of key methods: Covers algebraic equations (e.g., physics-informed neural networks), logic rules (e.g., constraint-based learning), and simulation results (e.g., surrogate modeling).

Results

The paper does not present new experimental results but provides a qualitative analysis of the surveyed literature. It identifies that the most common integration point is the loss function (via regularization), and that symbolic representations (logic, equations) are most frequently used. The taxonomy reveals gaps, such as under-explored combinations of knowledge sources and integration strategies.

Significance

This survey has become a highly cited reference (877 citations) for researchers working on data-efficient learning, neuro-symbolic AI, and scientific machine learning. It provides a common vocabulary and classification that facilitates cross-disciplinary collaboration. For practitioners, it offers a decision framework: given a domain's available prior knowledge, which representation and integration method is most suitable. The work has influenced subsequent research in physics-informed learning, knowledge-guided ML, and explainable AI.

knowledge-graphsprior-knowledgetaxonomysurveymachine-learninginformed-machine-learningknowledge-representation

Related Papers