ºÚÁϳԹÏ×ÊÔ´

Ana Ozaki is an associate professor at the University of Bergen, Norway. Her research area is Artificial Intelligence (AI). She is an AI researcher in the field of knowledge representation and reasoning and in learning theory.

Ozaki is interested in the formalisation of the learning phenomenon so that questions involving learnability, complexity, and reducibility can be systematically investigated and understood. Her research focuses on learning logical theories formulated in description logic and related formalisms for knowledge representation.

She is a member of the editorial boards of the and the . She has recently worked as for the 27th International Symposium on Temporal Representation and Reasoning.

Ozaki is fascinated by learning and reasoning processes and how they interact.

There is a mode of learning which is actually how we learn very often but that is not the mode of learning we refer to when we talk about machine learning. We learn very often by posing queries (questions). Our intuition is that by obtaining answers to our questions from a teacher we can learn faster and better (more accurately) than by randomly selecting learning material. In learning theory, one can show that learners can efficiently exactly identify an unknown target concept formulated e.g. as a deterministic finite automaton, a decision tree, or a set of Horn rules if they can pose queries to a teacher.

Ozaki is currently working with her team on strategies to learn Horn rules from neural networks by posing them queries. She believes that exciting and recent advances in machine learning need to be complemented with theoretical development so that systems can provide formal guarantees of classification results and become trustable. More information: 

News:

AAAI 2021

IJCAI 2020

Women in AI

IJCAI 2015

Recent Talks:

• Querying Neural Networks. , 2021.

• Learning Description Logic Ontologies: Five Approaches. CAIR, 2020.

• Learning Description Logic Ontologies. , 2020.

  • I was invited to give a talk at the Norwegian Artificial Intelligence Research Consortium (NORA).

• On the Complexity of Learning Description Logic Ontologies. , 2020.

  • I was invited to be one of the lecturers of the 16^th edition of the RW summer school, held online this year. It had more than 100 participants.

• Learning Description Logic Ontologies. , 2020.

  • I was invited to give a talk at the . This is a research unit associated with the CNRS (UMR 5800), the University of Bordeaux and the Bordeaux INP.

• Learning Ontologies: A Question-Answer Game. , 2019.

  • I was the keynote of the 7th edition of the workshop, co-located with IJCAI.

• Learning Ontologies: A Question-Answer Game. , 2019.

  • is one of the world’s premier meeting centers for informatics research.

• . Seminar om forsking, innovasjon og teknologi, 2019.

  • I was invited to give a talk at Media City Bergen.

Current PhD Students

Raoul Koudijs (PhD Student)

• • Research Topic: Learning with Membership Queries 

  • Institution: University of Bergen

  • Year: 2022-2025  

Grunde Wesenberg (PhD Student)

• • Research Topic: Traffic Prediction with Graph Neural Networks 

  • Institution: University of Bergen & Transportation Research Institute

  • Year: 2021-2026 (part-time student) 

Victor Lacerda (PhD Student)

• • Research Topic: Differential Learning in Description Logics 

  • Institution: University of Bergen

  • Year: 2021-2024  

Current Master Students

Emil Poiesz (Master Student)

• Research Topic: On Learning to Reason with BERT 

• Institution: University of Bergen 

• Year: 2023/2024

Sophie Blum (Master Student)

• Research Topic: Extracting Rules from Language Models via Queries 

• Institution: University of Bergen 

• Year: 2022/2023

Kristoffer Ã†søy (Master Student)

• Research Topic: Language to Rule Translation using Transformers 

• Institution: University of Bergen 

• Year: 2022/2023

Anders Imenes (Master Student)

• Research Topic: Query Answering in DL-Lite over Knowledge Graph Embeddings 

• Institution: University of Bergen 

• Year: 2021/2023

Former PhD Students 

Cosimo Damiano Persia (PhD Student)

• • Research Topic: Learning Possibilistic Logic Theories 

  • Institution: University of Bergen

  • Year: 2020-2023  

 (PhD Student)

• Research Topic: Temporal Logic over Finite Traces 

• Institution: Free University of Bozen-Bolzano 

• Year: 2018-2022

Former Master Students

Anum Rehman (Master Student)

• Research Topic: Finding Common Grounds: The Moral Machine Case 

• Institution: University of Bergen 

• Year: 2021/2023

Emilia Przybysz (Master Student)

• Research Topic: Verificatino of Tsetlin Machines 

• Institution: University of Bergen 

• Year: 2021/2022

Senai Askale (Master Student)

• Research Topic: Traffic Prediction with Tsetlin Machines 

• Institution: University of Bergen 

• Year: 2021/2022

Johanna Jøsang (Master Student)

• Research Topic: Mining Rules from Knowledge Graph Embeddings 

• Institution: University of Bergen 

• Year: 2021/2022

Cosimo Damiano Persia (Master Student)

• • Research Topic: Learning Query Inseparable Ontologies 

  • Institution: Free University of Bozen-Bolzano

  • Year: 2018/2019  

Ricardo Duarte (Master Student)

• • Thesis Title: Exact Learning of EL Ontologies 

  • Institution: Dresden University of Technology 

  • Year: 2017/2018

Topics for Master Students.

Neural Network Verification

Neural networks have been applied in many areas. However, any method based on generalizations may fail and this is by design. The question is how to deal with such failures. To limit them, one can define rules that a neural network should follow and devise strategies to verify whether the rules are obeyed. The main tasks of this project are to study an algorithm for learning rules formulated in propositional Horn, implement the algorithm, and apply it to verify neural networks.       

References:

Queries and Concept Learning by Angluin (Machine Learning 1988)

Exact Learning: On the Boundary between Horn and CNF by Hermo and Ozaki (ACM TOCT 2020).

Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples by Weiss, Goldberg, Yahav (ICML 2018)

Knowledge Graph Embeddings

Knowledge graphs can be understood as labelled graphs whose nodes and edges are enriched with meta-knowledge, such as temporal validity, geographic coordinates, and provenance. Recent research in machine learning attempts to complete (or predict) facts in a knowledge graph by embedding entities and relations in low-dimensional vector spaces. The main tasks of this project are to study knowledge graph embeddings, study ways of integrating temporal validity in the geometrical model of a knowledge graph, implement and perform tests with an embedding that represents the temporal evolution of entities using their vector representations.

References:

Translating Embeddings for Modeling Multi-relational Data by Bordes, Usunier, Garcia-Durán (NeurIPS 2013)

Temporally Attributed Description Logics by Ozaki, Krötzsch, Rudolph (Book chapter: Description Logic, Theory Combination, and All That 2019)

Attributed Description Logics: Reasoning on Knowledge Graphs by Krötzsch, Marx, Ozaki, Thost (ISWC 2017)

Decidability and Complexity of Learning 

Gödel showed in 1931 that, essentially, there is no consistent and complete set of axioms that is capable of modelling traditional arithmetic operations. Recently, Ben-David et al. defined a general learning model and showed that learnability in this model may not be provable using the standard axioms of mathematics. The main tasks of this project are to study Gödel's incompleteness theorems, the connection between these theorems and the theory of machine learning, and to investigate learnability and complexity classes in the PAC and the exact learning models.

References:

Learnability can be undecidable by Ben-David, Hrubeš, Moran, Shpilka, Yehudayoff (Nature 2019)

On the Complexity of Learning Description Logic Ontologies by Ozaki (RW 2020)

Learning Ontologies via Queries

In artificial intelligence, ontologies have been used to represent knowledge about a domain of interest in a machine-processable format. However, designing and maintaining ontologies is an expensive process that often requires the interaction between ontology engineers and domain experts. The main tasks of this project are to study an algorithm for learning ontologies formulated in the ELH description logic, implement the algorithm, and evaluate it using an artificial oracle developed in the literature that simulates the domain expert.

References:

Learning Query Inseparable ELH ontologies by Ozaki, Persia, Mazzullo (AAAI 2020)ExactLearner: A Tool for Exact Learning of EL Ontologies by Duarte, Konev, Ozaki (KR 2018)

Exact Learning of Lightweight Description Logic Ontologies by Konev, Lutz, Ozaki, Wolter (JMLR 2018)

Binarized Neural Networks

Binarized neural networks (BNNs) have recently attracted a lot of attention in the AI research community as a memory-efficient alternative to classical deep neural network models. In 2018, Narodytska et al. proposed an exact translation of BNNs into propositional logic. Using this translation, various properties such as robustness against adversarial attacks can be proved. The main tasks in this project are to study BNNs and the translation into propositional logic, implement an optimised version of the translation, and perform experiments verifying its correctness.

References:

Binarized neural networks by Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv,Yoshua Bengio (NeurIPS-16)

Verifying Properties of Binarized Deep Neural Networks by Nina Narodytska, Shiva PrasadKasiviswanathan, Leonid Ryzhyk, Mooly Sagiv, Toby Walsh (AAAI-18)

Machine Ethics

Autonomous systems, such as self-driving cars, need to behave according to the environment in which they are embedded. However, ethical and moral behaviour is not universal and it is often the case that the underlying behaviour norms change among countries or groups of countries and a compromise among such differences needs to be considered.

The moral machines experiment () exposed people to a series of moral dilemmas and asked people what should an autonomous vehicle do in each of the given situations. Researchers then tried to find similarities between the answers from the same region.

The main tasks of this project are to study the moral machine experiment, study and implement an algorithm for building compromises among different regions (or even people). We have developed a compromise building algorithm that works on behavioural norms represented as Horn clauses. Assume that each choice example from the moral machines experiment is behavioural norm represented as a Horn clause. The compromise algorithm is applied to these choices obtained from different people during the moral machines experiment. One of the goals of this project would be to determine how to (efficiently) compute compromises for groups of countries (e.g., the Nordic Countries and Scandinavia).

References:

The Moral Machine experiment by Edmond Awad, Sohan Dsouza, Richard Kim, Jonathan Schulz, Joseph Henrich, Azim Shariff, Jean-François Bonnefon, and Iyad Rahwan (Nature 2018)

Advisors: ,

Own topic combining logic and learning

Contact: Ana Ozaki  

My record track, from 2014 until now, is of 1 book chapter and over 40 peer-reviewed journal, conference, and workshop publications. Most publications have the author list in alphabetical order.

DBLP Link: https://dblp.org/pers/o/Ozaki:Ana.html

Ana Ozaki is the principal investigator of the project funded by RCN. The goal of this project is to study and develop new automated strategies for building ontologies. Ontologies can be understood as an unambiguous way of representing knowledge. The project has two main objectives. The first one is to extract knowledge from neural network models (NNs) by posing them queries. In this way, one can discover hidden rules encoded in NNs and represent them as an ontology. The second objective is to design ontology languages that approximate the expressivity of NNs and learn ontologies formulated in these enriched languages. As part of this project, Ozaki co-organized the research school  and is currently organizing the 37th International Workshop on Description Logic .

She is the leader of the Machine Learning Theory Work Package for the project "Machine learning for computational efficient predictions of long-term congestion patterns in large-scale transport systems" funded by RCN. The main goal of the project is to develop an AI-based sketch planning model that can reliably and quickly predict congestion patterns in complex transport systems for the present and future scenarios, varying traffic growth and road capacity.

She is the main organizer of the Logic and Learning reading sessions . Through the , she is currently working to expand her network of collaborators and to support research projects in her team.

Ana Ozaki has worked as the principal investigator of the project funded by Unibz. She has also collaborated with Montserrat Hermo within the project .