Storyboard

Perception is not a passive copy of the world, but an active inference: the nervous system combines sensory information (likelihood P(O|E)) with prior knowledge (prior P(E)) to estimate the state of the world P(E|O) using Bayes' rule. This Bayesian view explains illusions, perceptual constancy, and context effects: the brain uses expectations to interpret ambiguous data.

The optimal fusion of two independent estimators produces a lower variance than either of the two individual sources: ²_fus < min(², ²). This explains why two eyes offer better acuity than one, why binaural hearing localizes better than monaural hearing, and why combining vision and proprioception improves balance. The brain roughly implements this minimum variance estimator.

The generative model g(E,) describes how the brain predicts what it should see/hear/feel given a hypothetical state of the world. Perception according to Friston's 'free energy principle' is the continuous minimization of the discrepancy between predictions and observations. 'Prediction error' neurons in the cortex encode this O - g(E,) difference.

The Kalman filter is the optimal linear estimator for dynamical systems with Gaussian noise. The Kalman gain K_t automatically weights how much to trust the new observation z_t versus the previous prediction x_t: if sensory noise is low (reliable measurement), K_t 1 and the observation is trusted; if the noise is high, K_t 0 and the prediction is confident. The cerebrum implements similar mechanisms for motor control.

These statistical tools explain phenomena such as perceptual constancy (the brain automatically corrects lighting to see constant colors), sensory capture (the most reliable modality dominates fused perception), and 'rubber hand illusion' effects where the brain updates its body model in the face of coherent multisensory evidence.

>Model

ID:('ky', 572)

Description

ID:(0, 572)