Paper on cooperative reasoning accepted at ECML PKDD 2026

A paper by the research group of Prof. Christian Bartelt has been accepted at the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2026. The conference, which is ranked CORE A, is one of the leading international venues for research in machine learning, artificial intelligence, and knowledge discovery from data. This year’s edition will take place in Naples, Italy, from September 7 to 11, 2026.

The accepted paper, titled “Beyond Either-Or Reasoning: Transduction and Induction as Cooperative Problem-Solving Paradigms,” investigates how two different forms of reasoning can work together to solve Programming-by-Example tasks. In these tasks, a system is given examples of desired input-output behavior and must infer how to produce the correct outputs for new cases.

Traditionally, such problems are approached either through induction, where the system learns a general rule or executable program from examples, or through transduction, where the system directly predicts outputs without constructing an explicit program. Existing methods often treat these paradigms separately or combine them in a fixed hierarchy, where one form of reasoning depends on the other. This can make it difficult for systems to recover when early decisions are wrong.

The paper proposes a different perspective: induction and transduction should not be viewed as competing alternatives, but as cooperative problem-solving paradigms. To explore this idea, the authors introduce TIIPS, short for Transductively Informed Inductive Program Synthesis. TIIPS uses transductive reasoning to guide the search for programs while still preserving the independence and strengths of inductive program synthesis.

Across three standard Programming-by-Example domains, TIIPS consistently outperforms state-of-the-art baselines and produces solutions that more closely match the intended program behavior. The results suggest that cooperative reasoning can be a promising direction for combining symbolic and neural approaches to artificial intelligence.

The full paper is available on arXiv.