Location-Specific and Location-Invariant Neural Representations of Visual Summary Statistics

Abstract

Human vision efficiently navigates information-dense environments by extracting summary statistics—the average properties of item groups—to circumvent capacity limits. However, the neural transition from location-specific sensory registration to location-invariant abstract representation remains poorly understood. We recorded high-density EEG while participants of either sex performed an ensemble size discrimination task, using time-resolved multivariate pattern analysis and cross-visual-field generalization to dissociate these two levels of representation. Our results reveal a clear temporal hierarchy: location-specific ensemble size information emerged as early as ∼40 ms poststimulus, significantly preceding the onset of location-generalized, abstract representation at ∼89 ms. During the early location-specific decoding phase, we observed a distinct left-visual-field advantage, with higher neural decoding accuracy predicting superior behavioral precision. Critically, error trials were characterized by premature neural generalization, suggesting an inherent trade-off: while abstraction is essential for efficient summarization, sacrificing sensory fidelity too early impairs perceptual accuracy. Furthermore, distinct oscillatory mechanisms supported this transformation—low-frequency (delta/theta) activity underpinned location-specific encoding, whereas mid-frequency (alpha/beta) oscillations robustly sustained location-invariant abstraction. Individual differences in the strength of these late-stage location-specific and location-invariant representations were predictable from intrinsic resting-state occipitoparietal gamma power. Together, these findings provide a comprehensive neural model of ensemble perception, demonstrating that the brain constructs abstract summary statistics through a temporally ordered, spectrally specific transformation that balances sensory fidelity with representational abstraction.