Seminar iz veštačke inteligencije, 27. oktobar 2021.

Naredni sastanak Seminara biće održan onlajn u sredu, 27. oktobra 2021. od 19 do 20 časova.

Predavač: Miloš Simić


Apstrakt: The traditional binary classification framework constructs classifiers that may have good accuracy but whose false positive and false negative error rates are not under users’ control. In many cases, one of the errors is more severe, and only the classifiers with the corresponding rate lower than the predefined threshold are acceptable. This lecture will present a way to combine binary classification with statistical hypothesis testing to control the target error rate of already trained classifiers.
In particular, we will show how to turn binary classifiers into statistical tests, calculate the classification p­-values, and use them to limit the target error rate. We will illustrate the approach on two classifiers.
The first is a neural network trained to detect normal distributions by inspecting small samples drawn from them. The second is an SVM for risky loan detection. The approach has a very high potential for use in the everyday practice of data analysis and machine learning in both science and industry.

Registraciona forma za učešće na Seminaru je dostupna na:

Prenos seminara je dostupan i registrovanim i neregistrovanim korisnicima na sledećem linku:

Ostavite vaš komentar:

(nece biti prikazano)