Continual learning in neural networks: Addressing catastrophic forgetting through scalable and robust methods

Abstract

Continual Learning (CL) enables artificial neural networks to adapt to evolving task distributions while retaining previously acquired knowledge—a critical requirement for real-world applications such as robotics, personalized healthcare, and autonomous systems. This survey presents a comprehensive and in-depth analysis of CL paradigms, theoretical foundations, evaluation methodologies, and recent advancements. We introduce a refined taxonomy of CL approaches, compare state-of-the-art methods using quantitative benchmarks, and explore biologically inspired mechanisms. Finally, we propose forward-looking research directions to build robust, generalizable, and task-agnostic continual learning systems.

Ähnliche Arbeiten