Stochasticity

• I think that most neural phenomenon can be modelled deterministically.
• However, it is often simpler to use some form of randomness.
• For instance, you can specify the size of LIF neurons, to get different behaviour.
• You could more easily use in built random functions to get similar behaviour.
• Using randomness can often improve performance. If neurons have the same parameters and inputs, they behave exactly the same.
• With machine learning, randomness is often essential. (E.g. random initial weights in MLPs.)
• The simplified (and very simplified) neural and synaptic models we use can benefit substantially from randomness.