Answers are grounded in this paper and cite their sources.
Detecting novelty remains a persistent weakness of modern artificial intelligence (AI): systems that achieve impressive in-distribution accuracy can still fail under distribution shift, open-set conditions, and embodied settings where self-generated sensory changes confound external events. Nervous systems face closely related problems under tight constraints of latency, memory, and continual adaptation, suggesting that brain-inspired mechanisms may offer useful design principles rather than metaphors. This article reviews progress on two computational scenarios linking cerebellum and cerebellum-like circuits to novelty-related functions. The first, prediction–cancellation, learns sensory expectations from motor commands, state, context, and multimodal feedback, and subtracts predictable components via negative images, adaptive filtering, or internal forward models so that unpredicted external inputs emerge as residual signals. The second, high-dimensional expansion, transforms inputs into sparse, high-dimensional codes that alter similarity structure, reduce interference, and increase capacity; novelty and familiarity—along with set membership and frequency—can then be computed by plastic readouts. Although both scenarios can yield novelty-related signals, they define “novelty” differently and depend on different learning targets. Studies of the cerebellum proper most directly support prediction and residual coding of unexpected self-motion, whereas explicit memory-based novelty and familiarity readouts have been demonstrated most clearly in the Drosophila mushroom body. Finally, we connect these circuit-level ideas to recent AI algorithms and propose an explicitly conceptual model that integrates expansion coding with multi-timescale plasticity to support fast detection, short-term adaptation, and longer-term incorporation of persistent changes.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: