Leak and Fatigue

• Two other properties that can be included are leak and fatigue.
• And, another modelling issue is continous vs. discrete time.
• Neurons behave in real continuous time.
• You can approximate this to an arbitrary degree in discrete time, however, the granularity affects things. (Tighter grain is more accurate but more computationally expensive.
• If you look at discrete time, what do you do with activity that doesn't take you over the threshold at each cycle?
• One answer is to just throw it away, and a second is to keep it all.
• What really happens (biologically) is some leaks away. This makes it easier to fire later.
• Note that a neuron loses all activation when it does fire.
• A lot of models use leak.
• Another problem is that neurons become harder to fire once fired.
• This is modelled by fatigue.
• I'm rather unique in using FLIF models.