At present, there are neural networks in all aspects of engineering application, and younger brother is now learning neural network, a little conjecture.
Most of the current neural network is to adjust their own weights, so as to learn. Under the structure of a certain neural network, improve the convergence speed by improving the algorithm, or deepen the layer number, improve the performance. In the network learning process, the network itself just adjusts its weight, this stage is similar to the human learning process. Networks of different structures can be regarded as different levels of biology. For example, a single hidden layer of the BP network, is a relatively low-level organisms, the implicit derivative of a more network, such as convolutional neural network CNN, can be regarded as a more advanced species. When the low-level biological development of a certain stage, there will be evolution, in this analogy, the future may also appear a network, the structure is random initialization, the weights are random initialization, the authority value learning to a peak state, will automatically adjust their own network structure, simulating the process of biological evolution. After adjusting the structure, then continue to adjust weights, to learn. Once again, when learning a peak state, the structure will be adjusted again and again. If there is such an algorithm, the artificial neural network can evolve autonomously, that artificial intelligence may really become a kind of beyond human existence.
Just a few guesses when learning, welcome to shoot bricks.
A little conjecture about the neural network