Neuromorphic & Continual Learning
Spiking neural networks, forward-only learning, and learning from non-stationary data streams with concept drift.
Real-world data arrive continuously and change over time, and energy budgets are limited. This line studies brain-inspired and energy-efficient learning paradigms that adapt online.
It developed in several stages:
- Stream learning and concept drift, the subject of Jesús L. Lobo’s doctoral thesis (Extraordinary PhD Award): drift detection and adaptation, diversity-based ensembles, cellular automata as interpretable stream learners and lightweight hyper-parameter tuning. Izaskun Oregi’s thesis added online elastic similarity measures for streaming time series.
- Learning under extreme verification latency, where labels arrive very late or never, in María Arostegi’s thesis.
- Spiking neural networks: evolving SNNs for online learning, traffic forecasting and air pollution prediction, and SNNs for energy-efficient object detection and image restoration.
- Randomization-based neural networks (echo state networks, random vector functional link networks and extreme learning machines) as fast and frugal learners.
- Forward-only learning (the Forward-Forward algorithm and its variants), the most recent work: a biologically plausible alternative to backpropagation that suits continual learning and neuromorphic hardware, studied in Erik B. Terres-Escudero’s thesis.
Main contributions
- Spiking neural networks: an overview of SNNs and online learning (Lobo et al., 2020), evolving SNNs for drifting data streams and drift detection (Lobo et al., 2018; Lobo et al., 2020; Lobo et al., 2018), traffic and air pollution forecasting (Laña et al., 2019; Laña et al., 2018; Maciąg et al., 2023), energy-efficient object detection and image deraining (Martínez-Seras et al., 2023; Islam et al., 2026), OOD detection for SNNs (Martínez-Seras et al., 2023) and an overview of neuromorphic technology in Spain (Iturbe et al., 2024).
- Forward-only learning: the Forward-Forward algorithm with better generalization and stability through neural polarization (Terres-Escudero et al., 2024), a contrastive symmetric variant for continual learning (Terres-Escudero et al., 2024) and emerging neo-Hebbian dynamics with implications for neuromorphic computing (Terres-Escudero et al., 2026).
- Concept drift and stream learning: evolutionary diversity for drift adaptation (Lobo et al., 2018), cellular automata for drift detection and stream learning (Lobo et al., 2021; Lobo et al., 2021), lightweight hyper-parameter tuning (Lobo et al., 2021), imbalanced streams (Lobo et al., 2016), rank aggregation over non-stationary data (Irurozki et al., 2021), energy IoT systems (Lobo et al., 2020), drift and uncertainty (Lobo et al., 2023), and online elastic similarity measures for streaming time series (Oregi et al., 2017; Oregi et al., 2019; Oregi et al., 2022).
- Extreme verification latency: concept tracking without labels (Arostegi et al., 2018), the semi-supervised SLAYER approach (Arostegi et al., 2021) and the adaptive neural gas models AiGAS-dEVL (Arostegi et al., 2025; Arostegi et al., 2026).
- Randomization-based and reservoir computing: a critical review of randomization-based learning for renewable energy prediction (Del Ser et al., 2022), ensemble deep random vector functional link networks (Shi et al., 2022), evolutionary quantization of randomization-based networks (Del Ser et al., 2023), and echo state networks for traffic forecasting, trajectory prediction, fake news detection and driving behaviour classification (Del Ser et al., 2019; Manibardo et al., 2023; Del Ser et al., 2022; Del Ser et al., 2022).
Publications in this area (43)
All my publications related to this research line, in reverse chronological order, including those mentioned above.
2026
2025
2024
2023
- IEEE International Conference on Intelligent Transportation Systems (ITSC)
2022
2021
- European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)SLAYER: A Semi-Supervised Learning Approach for Drifting Data Streams Under Extreme Verification LatencyIn European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD), 2021