Postdoctoral researcher · ELLIS member
I develop machine learning methods for reasoning with structured knowledge, combining learned representations with explicit relationships to answer complex questions and support scientific discovery.
Currently I am a postdoctoral researcher at Eindhoven University of Technology (TU/e), working on knowledge graphs and multimodal foundation models for materials discovery within the SimuLingua Horizon Europe project.
I completed my PhD at Vrije Universiteit Amsterdam cum laude in 2024. Before joining TU/e, I held postdoctoral positions at Amsterdam UMC, working on knowledge graphs and rare disease research, and at VU Amsterdam in the Learning & Reasoning Group.
Research direction
How can we learn from structured knowledge
and reason with what we learn?
Knowledge graphs make relationships explicit, while learned representations help us work with incomplete information. I am interested in bringing these strengths together: systems that can combine evidence, follow relationships, and answer questions that require more than a single prediction.
My work has developed through projects in graph representation learning, complex query answering, and biomedical research. These experiences increasingly bring me to a shared question: how can structured knowledge support useful, interpretable reasoning in science? Materials discovery is the setting in which I am now exploring that question.
Research themes
Learning representations of structured knowledge
How can machine learning capture the information in graph-structured data? I study graph representation learning, including methods that combine graph structure with text and molecular data, scale learning to large graphs, and compute and explain similarity between entities.
Entities from text · BioBLP · GRAPES · Explaining node similarity
Complex reasoning with learned representations
How can we answer expressive queries over graphs when information is missing? I develop neural and neuro-symbolic query answering methods that use learned representations to reason over incomplete graphs and incorporate user preferences through soft constraints.
Applications to scientific discovery
How can graph learning and reasoning help scientists interpret data and investigate new hypotheses? I apply these methods to multimodal biomedical knowledge graphs in BioBLP, the interpretation of rare disease omics predictions, and materials discovery through knowledge graphs and multimodal foundation models in SimuLingua.
Recent Highlights
- Interactive Query Answering on Knowledge Graphs with Soft Entity Constraints has been accepted in Transactions on Machine Learning Research, extending query answering with constraints that are difficult to express in first-order logic.
- Explaining Graph Neural Networks for Node Similarity on Graphs has been accepted in Transactions on Machine Learning Research, studying how to explain learned notions of similarity in graph neural networks.
- As lead co-applicant, I helped secure computing time for Large Laboratory Models on national compute facilities via the NWO Large Compute Application program, with 300k CPU and 2M GPU compute credits.
Selected Publications
For a complete publication list, see Google Scholar.
*Indicates equal contribution.
Interactive Query Answering on Knowledge Graphs with Soft Entity Constraints
Combining graph queries with preferences that are hard to express as logical constraints.
Daniel Daza, Alberto Bernardi, Luca Costabello, Christophe Gueret, Masoud Mansoury, Michael Cochez, Martijn Schut.
Explaining Graph Neural Networks for Node Similarity on Graphs
Explaining which graph structures contribute to learned node similarity.
Daniel Daza, Cuong Xuan Chu, Trung-Kien Tran, Daria Stepanova, Michael Cochez, Paul Groth.
Counting Still Counts: Understanding Neural Complex Query Answering Through Query Relaxation
Examining neural query answering through simpler, relaxed queries.
Yannick Brunink*, Daniel Daza*, Yunjie He, Michael Cochez.
Select, Don’t Train: The Benefits of Modular Entity Disambiguation with LLM-Based Selection
Using language models within a modular approach to entity disambiguation.
F. Polat, Daniel Daza, P. Zhang, K. Zaporojets, P. Groth.
GRAPES: Learning to Sample Graphs for Scalable Graph Neural Networks
Learning which parts of a graph to sample for scalable representation learning.
Taraneh Younesian, Daniel Daza, Emile van Krieken, Thiviyan Thanapalasingam, Peter Bloem.
UnRavL: A Neuro-Symbolic Framework for Answering Graph Pattern Queries in Knowledge Graphs
Bringing learned predictions and symbolic structure together for graph pattern queries.
Tamara Cucumides, Daniel Daza, Pablo Barcelo, Michael Cochez, Floris Geerts, Juan L. Reutter, Miguel Romero Orth.
BioBLP: a modular framework for learning on multimodal biomedical knowledge graphs
Learning biomedical entity representations from graph structure and multiple data modalities.
Daniel Daza, D. Alivanistos, P. Mitra, T. Pijnenburg, M. Cochez, P. Groth.
Inductive Entity Representations from Text via Link Prediction
Connecting textual descriptions to representation learning on knowledge graphs.
Daniel Daza, M. Cochez, P. Groth.
Complex Query Answering with Neural Link Predictors
Answering complex logical queries by composing neural link predictions.
Erik Arakelyan*, Daniel Daza*, Pasquale Minervini*, Michael Cochez.
Teaching and Academic Service
I supervise bachelor’s and master’s research on machine learning for constructing graphs and learning representations over them.
At VU Amsterdam, I contributed to the design of the first edition of Machine Learning for Graphs, now a recurring MSc course, including a lecture, a homework assignment, and student working sessions. I have also taught lectures on graph representation learning, complex query answering, and learning with graphs.
I contributed to the CIKM 2023 tutorial Reasoning beyond Triples and review for venues including ICLR, ICML, NeurIPS, and TMLR. I was recognized as an ICML Gold Reviewer in 2026.
Download my CV for supervision, teaching, service, and the full publication record.