OCEANO 1 Lecture Notes - Lecture 33: Stochastic Gradient Descent, Fault Tolerance, Feature Vector
Document Summary
Principles of connectionist models (mimic neural computations) Computational operation in the brain: neurons passing on information to other sets of neurons. Large numbers of neurons need to perform computations in parallel for cognitive processing. Information is distributed across many neurons and connections. Connectionist approach aims to mimic neural computations. Not always fully accurate, since not all neural activity is fully understood. Memory connectionist models store a specific number of events before interference between memories changes the number and structure of them and their connections. Neuron receives excitatory/inhibitory signals from other neurons via synapses. If sum of inputs exceeds threshold, action potential is sent down the axon to the terminal where the signal is passed on to the dendrites of the next neuron via the synapse. Functional role of a unit in the model. Pass on information about activity of one set of units to another (analogous to neurons: neurons pass information about the level of their input.
