|
|
|
|
|
|
by leoh
2273 days ago
|
|
|
> Using real neurons avoids several other difficulties that software-based neural networks have. For instance, to get artificial neural networks to start learning well, their programmers usually have to engage in a laborious process of manually adjusting the initial coefficients, or weights, that will be applied to each type of data point the network processes. This seems silly. Once you've figured out these parameters once, it seems to me you can similate a single neuron reasonably well. Perhaps differences will emerge from networks. |
|
|
Not true for modern neural networks. We typically use random values with certain statistics and/or specialized schemes which depend on specific later details - but in any case it's a single function call and the defaults typically work well enough that it's an advanced topic.
The OP is probably referring to neural nets of old, before this recent explosion, where you had handfuls of perceptrons operating on very simple problems which would be trivial for modern ML.