Duet model unifies diverse neuroscience experimental findings on predictive coding

The brain continuously generates predictions about the external world. When stimulus X is presented repeatedly, the brain predicts that the next one is also X. A deviant stimulus Y elicits a stronger sensory response than the baseline, reflecting the amplification of an unexpected stimulus. Here, we introduce the duet predictive coding model, a minimal and biologically plausible framework in which neurons encode both positive and negative prediction errors. This model reproduces neural responses observed in vision and audition across diverse predictive coding paradigms, particularly omission. Our proposed circuit mechanism predicts (1) neurons tuned to negative prediction errors in the oddball paradigm, supported by experimental evidence in mice; (2) the magnitude of unexpected responses quantitatively depends on the dissimilarity between standard and deviant stimuli and diminishes with increasing interstimulus interval. Our findings suggest that the brain's deviance detection relies on dual-error computation, offering a unifying explanation across seemingly disparate experimental protocols.