David Tena Cucala
David Jaime Tena Cucala
Themes:
Interests
I work in the area of knowledge representation and reasoning in the field of artificial intelligence (AI).
The research question that guides my work is: what knowledge do intelligent systems possess? By "knowledge" I mean both factual information (what is there?, where is it?, etc) and models or theories that capture relevant aspects of how the world works. In what ways can we say that intelligent systems possess such knowledge? And how are they realised in practice?
Traditionally, the field of AI has approached this question by proposing system designs where "knowledge" is represented explicitly as a list of sentences of first-order logic. However, many intelligent systems (brains, for example) do not follow this approach and yet they show impressive capabilities. More recently, researchers have been analysing machine learning (ML) models to understand their computations in terms of human-understandable concepts.
My recent work looks at graph neural networks (GNNs) and tries to understand their computations in terms of logical rules that use human-understandable predicates. For example, my colleagues and I have identified a subclass of graph neural networks which are equivalent to tree-shaped Datalog programs, in the sense that both the GNNs and the logical programs implement exactly the same transformations on relational datasets. I am also interested to see whether a similar analysis can be applied to other ML models, such as transformers.
I have also worked on the design of reasoning algorithms for expressive Description Logics; in particular, I have worked on the reasoners PAGOdA and Sequoia, which support the logic behind the ontology language OWL 2 DL, a standard of the Semantic Web. Furthermore, I have also worked on developing logics for temporal reasoning, combning non-monotonic extensions of Datalog with Metric Temporal Logics.
For my broader interests, please see my personal website.
Biography
I studied Mathematics and Physics at Universitat Autònoma de Barcelona, from 2008 to 2013. I was originally interested in Astrophysics, but I became fascinated by metaphysical questions arising from modern physics. I applied for the MSt in Philosophy of Physics at University of Oxford in 2013, and I spent another year in the BPhil in Philosophy at the same university, studying Ethics and Philosophy of Science. My focus then shifted towards the question of how existing sources of knowledge can be aggregated to answer complex questions, like those of Philosophy. In 2015, I went on to study the MSc in Computer Science at the University of Oxford, with a focus on Knowledge Representation and Automated Reasoning. In 2016, I started a DPhil in Computer Science in the same area and research group, which I finished in 2020.
Selected Publications
-
Faithful Rule Extraction for Differentiable Rule Learning Models
Ian Horrocks Xiaxia Wang David J. Tena Cucala Bernardo Cuenca Grau
In The Twelfth International Conference on Learning Representations. 2024.
Details about Faithful Rule Extraction for Differentiable Rule Learning Models | BibTeX data for Faithful Rule Extraction for Differentiable Rule Learning Models | Link to Faithful Rule Extraction for Differentiable Rule Learning Models
-
Pay−as−you−go consequence−based reasoning for the description logic SROIQ
David Tena Cucala‚ Bernardo Cuenca Grau and Ian Horrocks
In Artif. Intell.. Vol. 298. Pages 103518. 2021.
Details about Pay−as−you−go consequence−based reasoning for the description logic SROIQ | BibTeX data for Pay−as−you−go consequence−based reasoning for the description logic SROIQ | DOI (10.1016/j.artint.2021.103518) | Link to Pay−as−you−go consequence−based reasoning for the description logic SROIQ
-
DatalogMTL with Negation Under Stable Models Semantics
Przemyslaw Andrzej Walega‚ David J. Tena Cucala‚ Egor V. Kostylev and Bernardo Cuenca Grau
In Meghyn Bienvenu‚ Gerhard Lakemeyer and Esra Erdem, editors, Proceedings of the 18th International Conference on Principles of Knowledge Representation and Reasoning‚ KR 2021‚ Online event‚ November 3−12‚ 2021. Pages 609–618. 2021.
Details about DatalogMTL with Negation Under Stable Models Semantics | BibTeX data for DatalogMTL with Negation Under Stable Models Semantics | DOI (10.24963/kr.2021/58) | Link to DatalogMTL with Negation Under Stable Models Semantics