Abstract
Cognitive functions in current artificial intelligence networks are tied to exponential increases in network scale, whereas the human brain can continuously learn hundreds of cognitive functions with remarkably low energy consumption. This advantage partly arises from the brain's cross-regional temporal development mechanisms, whereby the progressive formation, reorganization and pruning of connections from basic to advanced regions facilitate knowledge transfer and prevent network redundancy. Inspired by these, we propose the continual learning of multiple cognitive functions with a brain-inspired temporal development mechanism, enabling cognitive enhancement from simple to complex tasks in perception-motor-interaction (PMI). The model drives the sequential evolution of long-range inter-module connections to facilitate positive knowledge transfer and employs feedback-guided local inhibition and pruning to eliminate redundancies from prior tasks, thereby reducing energy consumption while preserving acquired knowledge. Experiments on the proposed cross-domain PMI dataset and on general datasets (CIFAR100 and ImageNet) demonstrate that the proposed method achieves continual learning capabilities while reducing network scale, without introducing regularization, replay or freezing strategies, and attains superior accuracy on new tasks compared with direct learning. The proposed method indicates that the brain's developmental mechanisms provide a valuable reference for exploring biologically plausible, low-energy enhancements of general cognitive abilities.