Our paper “Complex Query Answering with Neural Link Predictors” has received an Outstanding Paper Award.
Out of the 860 papers of the ICLR program, only 8 selected papers, that have been deemed of exceptional quality, have received this recognition! More information can be found in the official ICLR announcement. The actual paper can be found on openreview: https://openreview.net/pdf?id=Mos9F9kDwkz.
Congratulations Daniel Daza and Michael Cochez, and also to our collaborators in this work, from the UCL Center of Artificial Intelligence, Erik Arakelyan and Pasquale Minervini!
In this work, we show how to re-use models for 1-hop link prediction on knowledge graphs, to answer more complex queries involving larger sub-graphs. We improve upon previous methods that require orders of magnitude more training data.
The paper will be presented virtually on May 6 in the Outstanding Paper conference session 00:00 - 01:00 PDT, 03:00 - 04:00 EDT, 09:00 - 10:00 CET, 15:00 - 16:00 CST.
Invited paper at GKR 2020 (workshop at ECAI 2020) published 'Approximate Knowledge Graph Query Answering: From Ranking to Binary Classification'
Several members of the lab got invited to write a contributions to the proceedings of the GKR2020 workshop. In this paper, we propose an additional metric to be used for the evaluation of approximate knowledge graph query answering. We also propose a graph embedding model based on axis-aligbned hyperrectangles that seems weel suited for this task.
The paper is availble as an open access publication from the proceedings published in the Lecture Notes Artificial Intelligence series.
Paper accepted to IEEE Access - DeepKneeExplainer: Explainable Knee Osteoarthritis Diagnosis From Radiographs and Magnetic Resonance Imaging
In collaboration with researcher from Fraunhofer, Germany and ZB MED—Information Centre for Life Sciences, Cologne, Germany, Michael Cochez from the discovery lab got a paper accepted in IEEE Access. In this paper, we show the result of exhaustive experimentation with explainable neural network techniques applied on radiographs and Magnetic Resonance Imaging (MRI).
The paper is published as open access here: https://ieeexplore.ieee.org/document/9363889
IEEE Access: M. R. Karim, J. Jiao, T. Döhmen, M. Cochez, O. Beyan, D. Rebholz-Schuhmann and S. Decker, “DeepKneeExplainer: Explainable Knee Osteoarthritis Diagnosis From Radiographs and Magnetic Resonance Imaging,” in IEEE Access, vol. 9, pp. 39757-39780, 2021, doi: 10.1109/ACCESS.2021.3062493.
Rinke Hoekstra, Lead Architect at Elsevier, is Industry Director of Discovery Lab. Rinke was interviewed by ICAI to talk about the collaboration with the University of Amsterdam and Vrije Universiteit Amsterdam: ‘Within Elsevier this is already seen as one of the most successful collaborations with academic partners.’
Read the interview at the ICAI website.
Paper accepted at The Web Conference: 'Inductive Entity Representations from Text via Link Prediction'
Our paper, “Inductive Entity Representations from Text via Link Prediction”, has been accepted at The Web Conference, 2021. With Daniel Daza, Michael Cochez and Paul Groth, we investigate how to learn representations of entities in a knowledge graph given their textual description. We then reuse these representations in tasks of entity classification and information retrieval, obtaining significant improvements over previously proposed methods.
A preprint of this work can be found here: https://arxiv.org/abs/2010.03496
Our paper “Complex Query Answering with Neural Link Predictors” was accepted for oral presentation at ICLR 2021. It is the result of a collaboration with Erik Arakelyan and Pasquale Minervini from UCL. We show how to re-use models for 1-hop link prediction on knowledge graphs, to answer more complex queries involving larger sub-graphs. We improve upon previous methods that require orders of magnitude more training data.
The paper will be presented virtually in the first week of May.
The paper, titled “Message Passing Query Embedding” was accepted at the ICML 2020 Workshop on Graph Representation and Learning. It is authored by two of our lab members: Daniel Daza and Michael Cochez. The paper proposes a novel architecture for graph embeddings of knowledge graph queries, with important advantages compared to previous works.
Data scientists are developing a knowledge graph with researchers in mind in Elsevier’s DiscoveryLab, collaborating with Vrije Universiteit and University of Amsterdam.
Read the article on Elsevier Connect.