Backpropagation concept
Backpropagation is defined as one of the algorithms programmed in programming languages to train neural networks to detect the error and then sequentially return through the layers in the opposite direction to correct the error, and this process continues to iterate until there is no error. . False and the actual output matches the expected or requested output. The form is then approved as a trusted form.
A neural network consists of a set of layers, namely the input layer, the output layer, and a set of hidden layers in between. The introduction of the outputs, or in other words, their arrangement in the output layer.
The popular Google search engine is an example of a neural network. Where the image and word data represent the units in the input layer, while the search results shown to us in the search engine represent the units in the output layer.
The function of the backpropagation algorithm is to teach the neural network (the search engine) to display the units in the output layer (the search results) in the weight order assigned to the units in the input layer. Unit weights in the input layer until the actual results match the desired results.
Backpropagation mechanism
- The user enters the input, let's say he enters the value X.
- The input layer receives the value X.
- Inputs are given random weights and are denoted by the symbol W.
- The output of each neuron is computed from the input layer, through the hidden layers, to the output layer.
- The error is calculated on the output; Subtract the required output from the actual output.
- The weights are adjusted again to reduce the error, but vice versa, from the output layer, through the hidden layers, to the input layer.
- The process is repeated until the error value is as low as possible and the desired output is obtained.
The importance of back propagation
- Simple fast and easy to program
- It does not require the user to learn any special skills, he just needs to set the input numbers (weights).
- A standard approach works very efficiently, because the neural network distinguishes and organizes millions of data in record time, since this work requires a lot of time for humans to complete, in addition to many errors that are difficult to detect and modify as the algorithm does.
- Flexibility and possibility of continuous adjustment by readjusting the weights.
Backpropagation applications
- AI applications mainly use this algorithm. Among the applications that use this algorithm are:
- Artificial intelligence systems that recognize images and faces, for example, face printing on smartphones.
- Voice recognition systems, such as the Google voice search engine.
- Speaking of artificial intelligence systems, such as robots and smartphone assistants, in this case, the algorithm trains the neural network to pronounce letters and words.
Comments
Post a Comment