3 ms·
That fits well with neural nets, where noise plays a crucial role, from the random order of selecting examples into batches, to randomly dropping out synapses,
by visarga 8y ago
That fits well with neural nets, where noise plays a crucial role, from the random order of selecting examples into batches, to randomly dropping out synapses, directly injecting noise, or even starting prediction from pure noise (GANs). In RL the epsilon-greedy technique prescribes random actions from time to time, in order to better explore the environment. Noise, when used properly, makes neural nets better and more resilient.