Logic and Learning: an Algebra and Finite-Model-Theory Approach

Información del proyecto

LLAMA

Identificador del acuerdo de subvención: 101031081

DOI

10.3030/101031081

Proyecto cerrado

Fecha de la firma de la CE 26 Marzo 2021

Fecha de inicio 1 Octubre 2021

Fecha de finalización 15 Enero 2025

Financiado con arreglo a

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Coste total

€ 175 572,48

Aportación de la UE

€ 175 572,48

Coordinado por

UNIVERSITEIT VAN AMSTERDAM
Netherlands

Periodic Reporting for period 1 - LLAMA (Logic and Learning: an Algebra and Finite-Model-Theory Approach)

Período documentado: 2021-10-01 hasta 2023-09-30

Resumen del contexto y de los objetivos generales del proyecto

Computational learning theory is a branch of computer science that
studies the mathematical and algorithmic underpinnings of machine
learning. It provides the concepts and methods to classify the
computational feasibility of different learning problems. This
project lies at the intersection of computational learning theory
and logic It was concerned with techniques for learning concept
specified specified in logical languages such as first-order logic,
in the presence of background knowledge that is similarly specified
in logical languages such as first-order logic. The project builds
on connections between learning theory, universal algebra and
combinatorial graph theory in order to develop new learning algorithms
with applications in data management and knowledge representation.
Some of the main intended applications lie in the development of
new methodologies and systems for the interactive example-driven
specification and debugging of database queries and of description
logic concepts for knowledge representation.

Trabajo realizado desde el comienzo del proyecto hasta el final del período abarcado por el informe y los principales resultados hasta la fecha

The project has led to
- 10 publications at international conferences (IJCAI, PODS, ICDT, CSL, FOSSACS, ICDT, POPL, AIML)
- 5 journal publications (ACM Transactions on Database Systems, SIGMOD Record, Information Processsing Letters, ACM Transactions on Computational Logic, Logical Methods in Computer Science)
- 5 supervised MSc theses
- 4 keynote lectures at international conferences and workshops
- 2 one-week-long summer school courses
- 5 national or international workshops

The outcomes of the project include several novel learning algorithms,
as well as example-generation algorithms, for fragments of first-order
logic, including modal fragments, description logics, and classes
of database queries. In addition, limiting results are obtained
mapping out what classes of logical concepts are, or are not,
efficiently learnable from examples.

Furthermore, as part of the project, these technical results were applied
to develop a methodology for the interactive example-driven specification
of databae queries, and an implemented system for deriving description
logic concepts from data examples, for use in knowledge representation.

Avances que van más allá del estado de la técnica e impacto potencial esperado (incluida la repercusión socioeconómica y las implicaciones sociales más amplias del proyecto hasta la fecha)

The project results are expected to enable (i) new educational tools for use in teaching
information- and knowledge-management courses as well as foundational courses on
logic, and (ii) new example-driven tools to assist users in specifying, refining, and
debugging database queries, thereby enabling a wider class of users to perform correct
data analyses over structured data sources.

pic.jpg