Neural network
인공 신경망
A structure of layered units that multiply inputs by weights, add them up and pass the result on.
The name comes from the brain, but it is not a copy of one. Very simple multiply and add steps, stacked in great numbers, produce complex relationships.
Training measures how wrong the output is and pushes that error backwards to nudge the weights, a process called backpropagation.
The number of weights is the parameter count, which is what people mean by model size.
- BackpropagationSending the error backwards to adjust weights.
- SizeParameter count is not performance. Data and purpose matter more.