Course

ML - Machine Learnig


PDF Course Catalog Deutsche Version: ML

Version: 1 | Last Change: 23.09.2019 12:26 | Draft: 0 | Status: vom verantwortlichen Dozent freigegeben

Long name Machine Learnig
Approving CModule ML_BaTIN
Responsible
Prof. Dr. Lothar Thieling
Professor Fakultät IME
Level Bachelor
Semester in the year winter semester
Duration Semester
Hours in self-study 78
ECTS 5
Professors
Prof. Dr. Lothar Thieling
Professor Fakultät IME
Requirements basic skills in Java and/or C
basic skills in analysis and linear algebra
Language German
Separate final exam Yes
Literature
Géron, Aurélien, Hands-On Machine Learning with Scikit-Learn and TensorFlow, O'Reilly Medi
Final exam
Details
The students should demonstrate the following competences in an oral exam: 1.) Safe handling of basic concepts and mechanisms. 2.) Analyze problems in the field of machine learning and solve them with suitable methods. 3.) Analyze existing solutions and explain the used algorithmic and theory.
Minimum standard
At least 50% of the total number of points
Exam Type
The students should demonstrate the following competences in an oral exam: 1.) Safe handling of basic concepts and mechanisms. 2.) Analyze problems in the field of machine learning and solve them with suitable methods. 3.) Analyze existing solutions and explain the used algorithmic and theory.

Learning goals

Knowledge
fundamentals
types of learning
simple classifiers
simple predictors (Equalizer alias Linear Regression)
challenges in learning
linear regression as the simple predictor
linear regression as the simple classifier
training data (handling, analysis, processing)
gradient descent
quality measures
learning curve
multi-class classifier based on binary classifiers
multi-label-classification
logistic regression
simple neuronale Netze
the artificial neuron as a simple classifier
operation
activation function
bias
training a neuron
multi-layer-perceptron
operation
purposes of the layers
backpropagation training algorithm
tools for creating and training simple neural networks and handling training data
handling, analysis and preparation of training data
creating and configuring neural networks
training neural networks
verification of trained networks
Deep Neural Networks (DNNs)
basic problems
vanishing or exploding gradients
high training times
overfitting
solutions for the probblems mentioned above
appropriate initialization of the weights, non-saturating activation function, gradient clipping
accelerated optimization procedures, reuse of pre-trained layers
regularization to avoid overfitting
tools for creating and training DNNs
handling, analysis and preparation of training data
creating and configuring neural networks
training of neural networks
verification and validation trained networks
Convolutional Neural Networks (CNNs)
idea
architecture
convolutional layer
pooling layer
convolution as a basic operator for training and detection
architectures of CNNs for different problems
tools for implementation and training CNNs
Recurrent Neural Networks (RNNs)
idea
recurrent neurons
training of RNNs and Deep RNNs
Long Short Term Memory
architectures of RNNs for different problems
tools for implementation and training deep CNNs

Skills
the presented neural networks
specify
describe
evaluate the pros and cons

solving problems using tools
for handling, analysis and preparation of the training data
for implementationion, verification, validation and training of all neural presented networks
Expenditure classroom teaching
Type Attendance (h/Wk.)
Lecture 2
Exercises (whole course) 0
Exercises (shared course) 0
Tutorial (voluntary) 0
Special literature
keine/none
Special requirements
none
Accompanying material
lecture foils (electronic)

self-study tutorials for the tools
Separate exam
none

Learning goals

Skills
purposeful handling of the tools
deal with complex tasks in a small team
derive complex solutions that can be implemented using neural networks
Expenditure classroom teaching
Type Attendance (h/Wk.)
Practical training 2
Tutorial (voluntary) 0
Special literature
keine/none
Special requirements
none
Accompanying material
problem and task description (electronic)
tool chain for neural networks
self-study tutorials for the tools
Separate exam
none

© 2022 Technische Hochschule Köln