Lecture Introduction to Machine Learning and Data Analysis

Description

VAK NN
Category: Lecture+Lesson, 3 SWS
Master Course
ECTS: 6, Winter Semester
University of Bremen
Lecturer: PD Dr. Stefan Bosse

Course Content

  1. Sensors, digital sensor data, big data

    • Strongly and weakly correlated data (big data?)
    • Question about the models
    • Noise
  2. Basics of sensor data acquisition and processing

    • Sensor networks
    • Laboratory measurements
    • Non-destructive test methods - type of measurement data
    • Structure and load monitoring (SHM)
    • Data preprocessing (feature selection)
    • Principle Component Analysis (PCA)
  3. Basics of machine learning (metrics and taxonomy)

    • The functional approach: the black box model
    • Supervised learning - the expert is in demand!
    • Unsupervised learning, clustering - I see something you don't see?
    • Feedback-based learning - rewards lead to the goal!
    • Incremental learning - learning on data streams is a problem?
    • Agent-based and distributed learning - not here, but everywhere!
  4. Algorithms and models

    • Decision trees (C45, ID3, ICE), random forest trees - simple but good?
    • Support Vector Machines (binary and multi-classes) - The classic!
    • Artificial Neural Networks (Single and Multi-Layer) - Why No Deep Learning?
    • Regression method
    • Iterative, Randomized, and Evolutionary Learning Algorithms - Deterministic Models?
  5. Training, learning, prediction, test

    • Feature extraction - information from data
    • Flow diagrams - working instructions!
    • Test methods
    • Problems
    • Overfitting
    • Too much or too little data?
    • Quality of data, influence of noise on learning and prediction
  6. Applications, demonstrations, examples, laboratory exercises (integrated in 2-5)

Aims

  1. The students should understand the basics of machine learning and its tasks, goals, and applications, and gain insights into algorithms and data models. When deep learning, when and why not!

  2. The difference between the model (model representation and data structures) and the learning and prediction algorithms should be understood.

  3. The students should be able to use, distinguish and evaluate different learning methods on different training data in a simple way using simple laboratory exercises with a WEB-based ML kit and analysis tool (execution in the WEB browser or with node-webkit).

  4. Understanding and application of data preprocessing and importance of quantity and quality of the training data.

  5. An understanding of the problems in handling and using ML procedures is to be acquired using practical examples and exercises. The aim is to acquire the ability to independently select suitable ML methods for a specific problem from measurement and testing technology.

  6. At the end of the course, students should be able to process measurement data in a meaningful and targeted manner using ML procedures and be able to realistically evaluate the benefits and problems of using ML.