Noisy Target vs Target Distribution,Deterministic Target vs Deterministic Target Func

[ntvy95]

Hello, I am confused by the meaning of these four words:

Noisy Target vs Target Distribution

I understand that a target distribution is a probability distribution, but how about noisy target? What is a noisy target? A value or a function or a probability distribution?

Deterministic Target vs Deterministic Target Function

Same above, I understand that a deterministic target function is a function, but how about deterministic target? A value or a function?

And...

Noisy target = deterministic target f(x) = E(y|x) plus noise y − f(x)

Is y here a value or a function?

I am very confused...

Thank you very much for your time spending on here.

[magdon]

Apologies for the confusion.

Noisy target and deterministic target refer to a specific target value (for example at a specific point x). So you have a data point (x,y).

If y=f(x), we say that y is a deterministic target and f is a deterministic target function.

If for the same value of x you may observe different values of y (the example in the book: several people may have the same income but some may default on credit debt and some may not), then we say that y is a noisy target - its value is not deterministically determined given x. The way we model this situation is using a target distribution P[y|x]. So P[y|x] is the distribution of the (noisy) target, and we call it the target distribution.

Hope this helps.

Originally Posted by ntvy95

Hello, I am confused by the meaning of these four words:

Noisy Target vs Target Distribution

I understand that a target distribution is a probability distribution, but how about noisy target? What is a noisy target? A value or a function or a probability distribution?

Deterministic Target vs Deterministic Target Function

Same above, I understand that a deterministic target function is a function, but how about deterministic target? A value or a function?

And...



Is y here a value or a function?

I am very confused...

Thank you very much for your time spending on here.

[ntvy95]

Originally Posted by magdon

Apologies for the confusion.

Noisy target and deterministic target refer to a specific target value (for example at a specific point x). So you have a data point (x,y).

If y=f(x), we say that y is a deterministic target and f is a deterministic target function.

If for the same value of x you may observe different values of y (the example in the book: several people may have the same income but some may default on credit debt and some may not), then we say that y is a noisy target - its value is not deterministically determined given x. The way we model this situation is using a target distribution P[y|x]. So P[y|x] is the distribution of the (noisy) target, and we call it the target distribution.

Hope this helps.

So y here serves as a (random) variable?

[MaciekLeks]

Originally Posted by ntvy95

So y here serves as a (random) variable?

As the book says: "Instead of y=f(x), we can take the output y to be a random variable..."

[ntvy95]

Originally Posted by MaciekLeks

As the book says: "Instead of y=f(x), we can take the output y to be a random variable..."

Man, thank you. I have missed that important point. :'(