NeurIPS · 2026
Flow-based Spectral Kernel Learning for Nonstationary Attention
Learning richer relationships between tokens through nonstationary kernels and normalizing flows.
Coauthor · fourth authorRESEARCH / DARÍO C. LARESE
Better questions.
More expressive models.
I investigate how machine learning models represent relationships, uncertainty and time. My work connects attention mechanisms and probabilistic modeling with problems in healthcare and industry.
PEER-REVIEWED PUBLICATIONS
↗
IEEE Open Journal of Intelligent Transportation Systems · 2025
Transformer Vibration Forecasting for Advancing Rail Safety and Maintenance 4.0
Investigating transformers and spectral methods for forecasting the complex vibrations of railway axles.
First authorSpanish Journal of Psychiatry and Mental Health · 2026
Modeling Recurrent Suicide Attempts Using Probabilistic Hawkes Processes
A probabilistic approach to temporal event histories, with individualized intensity and survival curves.
First authorA RESEARCH CONVERSATION
Let me walk you through it.
FSKA · Darío C. Larese · Watch on YouTube ↗. Playing connects to YouTube; its privacy policy applies.
QUESTIONS IN PROGRESS
Work on
the horizon.
ONGOING RESEARCH
QKVAE: Variational Attention Mechanism for Transformers
Exploring a variational perspective on attention. This is ongoing research; no peer-reviewed publication is claimed here.
RESEARCH EXPLORATION
Coherent probabilities in decision models
Investigating whether different formulations of a decision question lead to coherent probability outputs.
SELECTED RESEARCH SOFTWARE
Kernel & flow-based attention ↗
Experiments with attention mechanisms, sequence classification and ablations. Research code, rather than a production library.
Neural Hawkes processes ↗
Feature-conditioned continuous-time event modeling, connected to the recurrent-event research.
Probability calibration ↗
Explorations of probability coherence across decision-model query formulations.
FSKA project website ↗
The comprehensive scientific companion: methods, results, presentation and reproducibility resources.
Academic home
Signal Processing Group (GTS), Department of Signal Theory and Communications, Universidad Carlos III de Madrid.