Learning
Biological learning systems
Biological brains learn continuously, from small data, without catastrophic forgetting. Replicating those properties is one of NeuroAI's hardest open problems.
Key takeaways
- Backpropagation is biologically implausible at scale; alternatives are an active research area.
- Continual learning, meta-learning, and few-shot learning all draw on biological inspiration.
- Hippocampal replay informs modern memory-augmented and experience-replay systems.
What biology does well
Continual lifelong learning, generalisation from sparse data, integration of new knowledge without overwriting old, and tight coupling between perception, memory, and action.
What AI is borrowing
Experience replay (from hippocampal consolidation), meta-learning (learning to learn), modular architectures, and emerging local learning rules inspired by cortical microcircuits.