Conclusion of Stochasticity Lecture

• Randomness should be used in neural nets (maybe not always).
• There are lots of ways of using randomness (e.g. connectivity, firing threshold, synaptic delay).
• Randomness needs to be accounted for, but...
• it also gives you lots of power.
• It allows the neurons to account for different behaviour and deal more robustly with inputs.
• Randomness actually a really important mechanism to understand to understand the behaviour of biological neural networks.
• Reading section 3.2.3 (simulating variability), section 5.2 (information transmission in random networks), and sections 10.3.2 and 10.3.3 (Boltzmann machines).