Types of Convergence in Machine Learning

[rifat28dddd]

Read also
Daniil Yugay: "Finally, my work coincides with my hobby. I am a data analyst"
Convergence in Machine Learning
The concept of convergence in machine learning is slightly different from convergence in classical mathematics, although the essence is similar. The difference is that the term is applied not to numerical sequences, but to the optimization of an ML model. Convergence means that the difference between the actual results and those predicted by the model has become minimal.

Here's how it works: During training, the models use two parameters, among others:

learning rate, or Learning Rate, is a hyperparameter kenya telegram data that shows how much the model needs to be changed after each error;
Errors, or Errors, are how much the model's results deviate from the truth.
Convergence is the point in learning after which there are fewer errors and the learning rate changes less. That is, it is a kind of “optimal” value that indicates that the model has achieved a high degree of accuracy.

In ML, convergence is divided into three groups. The division is based on what function was used to evaluate the learning process. Here are what these groups are and how they differ.

Convergence of the loss function. It describes how the values ​​of the error function change during the training of the algorithm.

When the function stops changing and reaches a final value, the algorithm is considered to have converged.
This type of convergence is a good illustration of the optimization process in a mathematical sense. But it does not always provide a complete picture of how well the model performs in terms of business results.
Convergence of the quality metric. In ML, other parameters besides error can be used. For example, precision or recall. For these metrics, convergence is also assessed — how the parameter changes as the model is trained.