Researchers are coming closer to describing exactly how knowledge is stored as patterns of connectivity in a neural network. They create from a framework "by implementing Markov chain Monte Carlo (MCMC) sampling in spiking networks of abstract model neurons." This shows how a neuron can draw an inference from a subset of the data available - a 'partial representation' - selecting from various possible interpretations. It's like: we see stripes in the jungle, is it shadows or a tiger, should we run? and we make a snap decision using this method by sampling the data. The full paper is very heavy in mathematics and neural network theory. The press release is a useful summary. Via Matthias Melcher.
Today: 0 Total: 3 [Share]
] [View full size