Neural network node self-increasing/decreasing method, device and storage medium
A technology of neural network and neural network model, applied in the field of device and storage medium, the self-increasing and decreasing method of neural network nodes, can solve the problem of slow neural network learning speed, increase the complexity of network structure, network can not have learning ability and information processing Ability to improve learning ability and reduce the amount of useless calculation
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0048] Embodiment 1 (increase and decrease neurons according to the threshold method)
[0049] The embodiment of the present invention provides a neural network node self-increasing method, which is applied to the training of the neural network model. The neural network node self-increasing method includes the following steps, as shown in the attached figure 1 Shown:
[0050] Step S11, designing the neural network structure according to requirements, and using the data to train the neural network model, so that the neural network model converges or the number of iterations of the neural network exceeds a certain threshold;
[0051] Step S12, performing increase and decrease operations on neurons layer by layer;
[0052] Mark the current neural network model as Mo, judge whether there are neurons that can be subtracted in the current layer, if there are neurons that can be subtracted in the current layer, then enter step S13, otherwise, enter step S15;
[0053] Among them, th...
Embodiment 2
[0066] Example 2 (increase or decrease neurons according to contribution)
[0067] The embodiment of the present invention provides a neural network node self-increasing method, which is applied to the training of the neural network model. The neural network node self-increasing method includes the following steps, as shown in the attached figure 2 Shown:
[0068]Step S21, designing the neural network structure according to requirements, and using the data to train the neural network model, so that the neural network model converges or the number of iterations of the neural network exceeds a certain threshold;
[0069] Step S22, performing increase and decrease operations on neurons layer by layer;
[0070] Mark the current neural network model as Mo, and judge whether there are neurons that can be subtracted in the current layer. If there are neurons that can be subtracted in the current layer, then enter step S23, otherwise, enter step S25;
[0071] The current layer is t...
Embodiment 3
[0082] Embodiment 3 (apparatus embodiment corresponding to the method)
[0083] In the embodiment of the present invention, a self-increasing and decreasing device for neural network nodes is also provided, which is applied to the training of neural network models, which marks the original neural network model as Mo, and performs layer-by-layer processing on the neural network model Mo Self-increasing operation processing; the self-increasing device includes:
[0084] The model analysis module is used to judge whether there are neurons that can be subtracted in the current layer, and if so, execute the subtraction operation processing module, otherwise execute the increase operation processing module;
[0085] The subtraction processing module is used to perform the subtraction operation on the neurons of the current layer, mark the neural network model after the subtraction operation as Mnew, and iteratively update the weights and biases of the associated neurons, said The a...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More