On Wednesday, April 22nd, two PhD students of CompCancer are giving a small tutorial on Neural Nets and Uncertaintity in Deep Learning.
Date: April 22nd
Time: 10 a.m.
Venue: Zoom
Speaker: Bettina Schmidt / Lorenz Rumberger
Uncertainty in deep learning (Lorenz Rumberger)
The tutorial gives an overview about methods for parameter and data related uncertainty quantification common in deep learning for regression and classification problems. After an introduction to the theoretical background, students will implement some of the discussed methods with tensorflow.
A short theoretical intro into neural nets (Bettina Schmidt)
Short derivation of mathematical formulation of simplest case of neural net (here: a classification) from logistic regression. Concept of activation function (relu) , loss / cost / objective function, backpropagation (problem of vanishing gradient), optizers (Adam), hyperparameteroptimization