作者: Gašper Tkačik , Olivier Marre , Dario Amodei , Elad Schneidman , William Bialek
DOI: 10.1371/JOURNAL.PCBI.1003408
关键词:
摘要: Maximum entropy models are the least structured probability distributions that exactly reproduce a chosen set of statistics measured in an interacting network. Here we use this principle to construct probabilistic which describe correlated spiking activity populations up 120 neurons salamander retina as it responds natural movies. Already groups small 10 neurons, interactions between spikes can no longer be regarded perturbations otherwise independent system; for 40 or more pairwise need supplemented by global interaction controls distribution synchrony population. show such “K-pairwise” models—being systematic extensions previously used Ising models—provide excellent account data. We explore properties neural vocabulary by: 1) estimating its entropy, constrains population's capacity represent visual information; 2) classifying patterns into metastable collective modes; 3) showing codeword ensembles extremely inhomogenous; 4) demonstrating state individual is highly predictable from rest population, allowing error correction.