What is retraining? And what is a model?

Question

Neural networks

What is retraining? And what is a model?
I can't understand at all. Just started learning machine learning.
Overfitting is when your algorithm has learned from the wrong data. Then I need to retrain the algorithm to more correct ones. Why then called retraining and not "incorrectly trained algorithm"?
The model, as I understand it, is not a physical value, such as a class, an algorithm. Is it some kind of connection? What synonym can you choose? Are there any examples?
Can you advise me to read something additional to understand?

Answer 1

What synonym can you choose?

Model - fr. modele, from lat. modulus - "measure, analogue, sample." In the context of machine learning, modeling is more in the direction of approximating the input data.
Our brains work the same way, you know. Based on his own experience, he builds his model of the surrounding world, and already on the basis of this model we determine what is what. Let's say if we once got burned in a fire, we knew approximately what the fire looks like, and the next time we see something that looks like fire, albeit a different color, for example, and without smoke, we will still understand that we don’t need to stick our hands in there , because the brain will run such a simulation within its model of the world and understand that the result is pain.
Overfitting Read wiki .

Answer 2

Firstly, they always train on the right data.
retraining (also called overfitting) is when exactly retrained, even on the "correct" data.
roughly speaking, this is when you trained the model in such a way that it simply repeats the data from the training sample exactly.
such an effect occurs when, for example, they try to fit the results of the algorithm to the training data as accurately as possible.
this is bad because your algorithm does not predict anything, but simply gives the answer exactly as in the learning data. no benefit from it.