https://www.dagstuhl.de/21192

09. – 12. Mai 2021, Dagstuhl-Seminar 21192

Approaches and Applications of Inductive Programming

Organisatoren

Andrew Cropper (University of Oxford, GB)
Luc De Raedt (KU Leuven, BE)
Richard Evans (DeepMind – London, GB)
Ute Schmid (Universität Bamberg, DE)

Auskunft zu diesem Dagstuhl-Seminar erteilt

Dagstuhl Service Team

Dokumente

Dagstuhl Report, Volume 11, Issue 4 Dagstuhl Report
Motivationstext
Teilnehmerliste
Gemeinsame Dokumente
Programm des Dagstuhl-Seminars [pdf]

Summary

The goal of Inductive Programming (IP) is to provide methods for induction of computer programs from data. Specifically, IP is the automated (or semi-automated) generation of a computer program from an incomplete information, such as input-output examples, demonstrations, or computation traces.IP offers powerful approaches to learning from relational data and to learning from observations in the context of autonomous intelligent agents. IP is a form of machine learning, because an IP system should perform better given more data (i.e. more examples or experience). However, in contrast to standard ML approaches, IP approaches typically only need a small number of training examples. Furthermore, induced hypotheses are typically represented as logic or functional programs, and can therefore be inspected by a human. In that sense, IP is a type of interpretable machine learning which goes beyond the expressivity of other approaches of rule learning such as decision tree algorithms. IP is also a form of program synthesis. It complements deductive and transformational approaches. When specific algorithm details are difficult to determine, IP can be used to generate candidate programs from either user-provided data, such as test cases, or from data automatically derived from a formal specification. Most relevant application areas of IP techniques is end-user programming and data wrangling.

This seminar has been the fifth in a series - building on seminars 13502, 15442, 17383, and 19202. In the wake of the recent interest in deep learning approaches, mostly for end-to-end learning, it has been recognized that for practical applications, especially in critical domains, data-intensive blackbox machine learning must be complemented with methods which can help to overcome problems with data quality, missing or errouneous labeling of training data, as well as providing transparency and comprehensibility of learned models. To address these requirements, on the one hand, explainable artificial intelligence (XAI) emerged as a new area of research and on the other hand, there is a new interest in bringing together learning and reasoning. These two areas of research are in the focus of the 2021 seminar. Futhermore, recent developments to scale up IP methods to be more applicable to complex real world domains has been taken into account. Based on outcomes of the fourth seminar (19202), the potential of IP as powerful approach for explainable artificial intelligence ("IP for XAI") has been be elaborated. Bringing together IP methods and deep learning approaches contributes to neural-symbolic intergration research. While two years ago (seminar 19202) focus has been on IP as interpretable surrogate model, in the 2021 seminar explainability of different addressees of explanations and their need to different types of explanations (e.g. verbal or example-based) are considered. For many real world applications, it is necessary to involve the human as teacher and judge for the machine learned models. Therefore, a further topic of the seminar has been to explore IP in the context of new approaches to interactive ML and their applications to automating data science and joint human-computer decision making.

Summary text license
  Creative Commons BY 4.0
  Andrew Cropper, Luc De Raedt, Richard Evans, and Ute Schmid

Dagstuhl-Seminar Series

Classification

  • Artificial Intelligence
  • Human-Computer Interaction
  • Machine Learning

Keywords

  • Interpretable Machine Learning
  • Explainable Artificial Intelligence
  • Interactive Learning
  • Human-like Computing
  • Inductive Logic Programming

Dokumentation

In der Reihe Dagstuhl Reports werden alle Dagstuhl-Seminare und Dagstuhl-Perspektiven-Workshops dokumentiert. Die Organisatoren stellen zusammen mit dem Collector des Seminars einen Bericht zusammen, der die Beiträge der Autoren zusammenfasst und um eine Zusammenfassung ergänzt.

 

Download Übersichtsflyer (PDF).

Publikationen

Es besteht weiterhin die Möglichkeit, eine umfassende Kollektion begutachteter Arbeiten in der Reihe Dagstuhl Follow-Ups zu publizieren.

Dagstuhl's Impact

Bitte informieren Sie uns, wenn eine Veröffentlichung ausgehend von
Ihrem Seminar entsteht. Derartige Veröffentlichungen werden von uns in der Rubrik Dagstuhl's Impact separat aufgelistet  und im Erdgeschoss der Bibliothek präsentiert.