How long does it take to train a neural network, neural network training takes too long

Why batch training neural networks

Does it require multiple trainings when reusing a neural network

Why does the accuracy of the rolling machine neural network decrease a lot after increasing the number of training sessions

How to understand "training" in multiple weight training in neural network

Neural network, 500 training samples, why is it better than 6000 training samples? After training, 500 predictions are better than 6000 samples!

It is not that the more training samples the better, it varies from subject to subject.1. The most important thing about the sample is the correctness and accuracy.The sample you choose must first correctly reflect the internal laws of the system process.

Many of the sample data we have taken from the production site may be bad samples, and such samples will interfere with your neural network training.Usually we think that bad samples are just individual phenomena, so we want to resist the negative effects caused by bad samples by making the sample size as large as possible.

2, followed by the balance of the sample data distribution.The samples you choose should preferably involve various situations that may occur in the system process, so that the regular characteristics of the system in various situations can be taken into account.

Usually we don't know much about the internal laws of the system, so we want to "carpet" all aspects of the object system with the largest possible sample size.3. Again, the size of the sample data, which is the question you have to ask.

In the case of ensuring the quality and distribution of the sample data, the size of the sample data determines the accuracy of your neural network training results.The larger the sample data volume, the higher the precision.

Because the sample size directly affects the computing time of the computer, if the accuracy meets the requirements, we do not need too much sample data, otherwise we will have to wait a long time for training.

Additionally, whether it is a radial basis (rbf) neural network or a classic bp neural network, it is only a specific training method. For enough iterations, the accuracy of the training results tends to be consistent. The methodIt only affects the convergence speed (operation time) of the calculation, and is not directly related to the sample size.

How to determine when the training set size is "large enough"?

The generalization ability of a neural network mainly depends on 3 factors: 1. The size of the training set 2. The architecture of the network 3. The complexity of the problem Once the architecture of the network is determined, the generalization ability depends on whether there is sufficientTraining set.

The appropriate number of training samples can be estimated using Widrow's rule of thumb.

The rule of thumb states that in order to obtain a good generalization ability, we need to meet the following conditions (Widrow and Stearns, 1985; Haykin, 2008): N=nw/e where N is the number of training samples, and nw is the number of training samples in the network.is the number of touch weights, and e is the network error allowed by the test.

So if we allow 10% error, we need about 10 times as many training samples as there are weights in the network.

Why support vector machines are better than neural networks?Can't neural networks be trained many times?