Fruit Fly-Inspired AI Learns New Skills Without Forgetting Old Ones

Inspired by the organization of the Drosophila learning and memory system, researchers propose a hierarchical modular principle for general continual learning — the ability to learn from sequential experience while retaining and adapting prior knowledge. Conventional continual learning is typically studied with offline task-wise training and clear task boundaries, leaving a gap from general continual learning under online, uncertain, and evolving data streams, where systems must separate conflicting experience to reduce interference while integrating compatible experience to promote generalization. The principle coordinates both functions through expert specialization and ensemble integration, instantiated as lightweight modular adaptation of pretrained foundation models that combines brain-inspired random expansion for expert routing with diversified modular integration across spatial and temporal scales. Across visual recognition, vision-language understanding, ego-exo video understanding, and embodied vision-language-action learning, the method consistently improved learning under online and uncertain data streams, with gains exceeding 50 percentage points over replay-free alternatives in embodied manipulation. The authors are Hongwei Yan and 9 other authors, and the paper is 50 pages.