Destroying synapses in a machine-learning model of the neocortex reduces the confidence of its predictions before reducing their number. The model, based on the temporal memory algorithm, was trained on random letter sequences representing affective states. Removing 50% of synapses only slightly lowered the number of predictions, but a 25% reduction distinctly dropped prediction confidence. This suggests that in major depressive disorder, synaptic loss in interoceptive cortices could trap the brain in limited affective states with high prediction error. The growth of new synapses, as proposed for ketamine and psilocybin, would allow more confident and futuristic predictions.
Synaptic atrophy in major depressive disorder may impair the brain's ability to confidently predict future affective states, even when predictions remain accurate. Using a temporal memory algorithm that mimics a single neocortical layer with Hebbian learning, researchers simulated depression by progressively destroying synapses. Destroying 50% of synapses slightly reduced the number of predictions, but a 25% reduction distinctly lowered prediction confidence. This suggests that in depression, interoceptive cortices become stuck in limited affective states with high prediction error. Treatments like ketamine and psilocybin may help by growing new synapses, enabling more confident and futuristic predictions.