Back to search

FRINATEK-Fri prosj.st. mat.,naturv.,tek

Learning Description Logic Ontologies

Alternative title: Læring ontologier

Awarded: NOK 7.9 mill.

Our world is becoming more and more digital. Digital platforms are not anymore a convenient alternative but a necessity that affects all organizations worldwide and generates large amounts of data. The larger datasets grow the more difficult it is to understand and abstract the relevant information from them. One of the most fundamental challenges in artificial intelligence (AI) is to automatically abstract such knowledge from the data and concisely represent it, so that it can be interpreted and explained in an accurate way. This project addresses the knowledge acquisition bottleneck in AI. The goal is to study and develop new automated strategies for acquiring interpretable knowledge, represented as an ontology, from large datasets. Ontologies can be understood as an unambiguous way of representing knowledge. In the field of knowledge representation, ontologies have been used to express knowledge about a domain of interest. The project contains two main strategies for learning ontologies. One of them is to design ontology languages that approximate the expressivity of neural network models (NNs) and learn ontologies formulated in these enriched languages. The second one is to extract knowledge from NNs by posing them queries. Algorithms that learn by posing queries traverse the search space in a systematic way, performing queries to identify the behaviour of the oracle, in this case, an NN. We propose the use of such algorithms to learn domain knowledge represented in an NN.

Our world is moving into a truly digital era. The transition into digital platforms is not anymore a convenient alternative but a necessity, that affects all organizations worldwide and generates large amounts of data. The larger datasets grow the more difficult it is to understand and abstract the relevant information from them. One of the most fundamental challenges in artificial intelligence (AI) is to automatically abstract such knowledge from the data and concisely represent it, so that it can be interpreted and explained in an accurate way. Our goal is to study and develop new automated strategies for building ontologies—which are explainable knowledge representation formalisms based on logic. Ontologies are useful to express the relevant knowledge about a domain of interest. This proposal is based on learning models from computational learning theory. The project contains two methods for learning ontologies. The first employs algorithms designed for the exact learning model and the second is a differentiable way of learning ontologies in the probably approximately correct learning model.

Publications from Cristin

No publications found

No publications found

No publications found

No publications found

Funding scheme:

FRINATEK-Fri prosj.st. mat.,naturv.,tek