Topic: Adaptive and Continual Learning for Intelligent Systems

Intelligent systems increasingly operate in dynamic and uncertain environments where data distributions evolve, new knowledge continuously emerges, and previously learned models may become outdated. In practice, deployment conditions rarely match those seen during training. Sensors may fail, become noisy, or be physically damaged (e.g., by stones or dirt). The set of sensors available in operation may differ from the one used for training. Knowledge transferred from a related domain may help or harm performance on the target task. Adaptive and continual learning provide promising solutions by enabling intelligent systems to continuously acquire, transfer, update, and retain knowledge while adapting to changing environments.

This Special Issue aims to bring together recent advances in adaptive learning, continual learning, and intelligent systems, covering novel theories, algorithms, architectures, evaluation methodologies, and real-world applications. We particularly welcome research addressing dynamic data, concept drift, limited or imbalanced data, knowledge retention, memorization and forgetting, transfer learning and negative transfer, model adaptation, and lifelong learning. We also encourage work on learning with missing or degraded sensors and on transferring knowledge across sensor configurations, especially in safety-critical applications such as autonomous driving.

Models that keep changing after deployment also raise questions of trust. Users, developers, and regulators need to know what has changed in a model, why it changed, and whether the system remains reliable, fair, and safe. Explainable AI (XAI) for continual learning remains underexplored: most explanation methods assume a static model, whereas a continual learner evolves, and its explanations may drift along with it. Contributions on trustworthy, credible, and explainable adaptive learning are therefore especially welcome.

Finally, reliably evaluating systems whose behavior changes over time is a research challenge in its own right. We invite contributions proposing robust evaluation protocols, metrics, and benchmarks for adaptive and continual learning systems.

Topics of interest include, but are not limited to:

· Foundations of adaptive and continual learning
● Continual, lifelong, and incremental learning;
● Adaptive and online learning;
● Catastrophic forgetting, knowledge retention, and the stability–plasticity trade-off;
● Adaptive memory, rehearsal, and knowledge management;
● Memorization in learning systems and its relation to generalization, forgetting, and privacy;
● Machine unlearning;
● Transfer learning and domain adaptation, including sim-to-real transfer;
● Transferability measures and negative transfer: estimation, detection, and mitigation.

· Learning from non-stationary and imperfect data
● Concept drift detection and adaptation;
● Learning from data streams, including delayed, partial, or noisy labels;
● Test-time and source-free adaptation;
● Open-world learning, novelty detection, and out-of-distribution detection;
● Active learning and self-labeling;
● Class-imbalanced and few-shot learning;
● Ensemble learning and adaptive ensembles.

· Multi-sensor and multimodal systems
● Multimodal and adaptive perception;
● Learning with missing, noisy, or degraded sensor data (e.g., due to sensor failure, occlusion, dirt, or physical damage) and graceful degradation of multi-sensor models;
● Knowledge transfer across sensor configurations, e.g., models trained on RGB and NIR data but deployed with RGB only, including learning using privileged information and cross-modal distillation.

· Trustworthy and explainable adaptive learning
● Explainable AI (XAI) for continual and adaptive learning, including explaining model updates, concept drift, and forgetting;
● Stability, consistency, and evaluation of explanations as models evolve over time;
● Explanation-guided adaptation, e.g., using explanations to detect drift, mitigate forgetting, or avoid negative transfer;
● Trustworthy and credible AI: reliability, uncertainty quantification, calibration, and fairness of adaptive systems;
● Safety, accountability, and human oversight of systems that adapt after deployment.

· Robustness and evaluation
● Robust evaluation of adaptive and continual learning systems: realistic protocols, metrics, benchmarks, and statistically sound comparison;
● Robust and efficient machine learning, including resource-constrained and on-device continual learning.

· Generative AI and applications
● Generative AI and adaptive learning, including continual adaptation of foundation models;
● Adaptive learning for robotics and autonomous systems;
● Autonomous driving and intelligent transportation;
● Applications in smart manufacturing, remote sensing, healthcare, IoT, and cyber-physical systems.