Backpropagation
Training algorithm that updates weights by spreading the error in the nervous network from the output.
My resemblance (backpropagation) is a training algorithm that calculates the difference between the forecast of a nervous network and the actual value (losing) and imposes the gradient from the output layer throughout the network to determine the source of this error. The mathematical basis is based on the chain rule on the rise: the derivative according to each weight of the total error indicates how much the weight contributes to the error. After calculating these derivatives (gradants), a optimization algorithm like gradyan landing will update up to a small step in the direction of reducing the error of each weight. The process gradually lowers the error rate of the network by applying again (in case of nookles) on all educational data.
The importance of my backdrop comes from the fact that millions of or even billions of weight in multi-layer (derin) networks can efficiently update; practical training of large-scale nerve networks without algorithm would not be possible. This method, popular in the 1980s, at the end of 2000s, fired the wick of deep learning revolution, coupled with GPU computing power and large data clusters. Today, libraries such as PyTorch and TensorFlow automate the resemblance to "automatic derivative" (autodiff) mechanisms, so researchers can design complex model architectures without having to write gradyan accounts manually.
The importance of my backdrop comes from the fact that millions of or even billions of weight in multi-layer (derin) networks can efficiently update; practical training of large-scale nerve networks without algorithm would not be possible. This method, popular in the 1980s, at the end of 2000s, fired the wick of deep learning revolution, coupled with GPU computing power and large data clusters. Today, libraries such as PyTorch and TensorFlow automate the resemblance to "automatic derivative" (autodiff) mechanisms, so researchers can design complex model architectures without having to write gradyan accounts manually.