Viktor Levandovskyy AA II, SS2020
Home
CV
Research
Teaching
(For) Students
Publications
Projects
Conferences
My SINGULAR
Songs

Angewandte Algebra II


(Mathematical Aspects of Neural Networks and Machine Learning)


Priv.-Doz. Dr. Viktor Levandovskyy

Information


All the information about this lecture is located in RWTH Moodle.
We meet on Wed and on Thu from 16:30 till 18:00 at a Zoom Meeting .
Register for lectures and example classes at RWTH online.
Through this registration you will gain access to the Moodle.

Contents

Sigma-Perceptron, separating hyperplane, Perceptron Learning Algorithm.
The unique separating hyperplane, Line Search Algorithm.
Optimal separation via quadratic optimization, Positive kernel, Support Vector Learning.
K-layer feed-forward networks and sigma-perceptrons. Realizations via k-layers.
Sigmoid functions, gradient descent, backpropagation algorithm.
Automatic differentiation and variants of backpropagation.
Discrete dynamical system, Lyapunov function, attraction, Hopfield network.
Theorems on fixed points in Hopfield networks, transient length and radius of attraction.
From Hebbian learning to projection learning.


University of Kassel Faculty10 Mathematik und NaturwissenschaftenInstitut für Mathematik