AI Engineering and Voice Assistants

Neural networks and deep learning approaches have revolutionized the fields of data science and artificial intelligence in recent years, and applications built on these techniques have reached or surpassed human performance in many specialized applications.

After a short review of the origins of neural networks and deep learning, this course will cover the most common neural network architectures and discuss in detail how neural networks are trained using dedicated data samples, avoiding common pitfalls such as overtraining.

The course includes a detailed overview of alternative methods to train neural networks and further network architectures which are relevant in a wide range of specialized application scenarios.

Kursinhalte

• Introduction to Neural Network and Deep Learning
  • The Biological Brain
  • Perceptron and Multi-Layer Perceptrons
• Network Architectures
  • Feed-Forward Networks
  • Convolutional Networks
  • Recurrent Networks, Memory Cells and LSTMs
• Neural Network Training
  • Weight Initialization and Transfer Function
  • Backpropagation and Gradient Descent
  • Regularization and Overtraining
• Alternative Training Methods
  • Attention
  • Feedback Alignment
  • Synthetic Gradients
  • Decoupled Network Interfaces
• Further Network Architectures
  • Generative Adversarial Networks
  • Autoencoders
  • Restricted Boltzmann Machines
  • Capsule Networks
  • Spiking Networks

Fakten zum Modul: Niveau: Master · Unterrichtssprache: English · 5 ECTS-Punkte · Modulprüfung: Mündliche Prüfung

Reinforcement learning allows computers to derive problem-solving strategies without being explicitly programmed for the specific task, similar to the way humans and animals learn.

After introducing the concepts of reinforcement learning, the course discusses the properties of Markov chains and single- and multi-armed bandits in detail. Special attention is given to the understanding of value functions and discounted value functions.

The course connects reinforcement learning with neural networks and deep learning and discusses how Q-Learning approaches can be used to utilize deep learning methods in reinforcement learning problems, including extensions such as double Q-Learning, hierarchical learning, and actor-critic learning.

Finally, the course discusses reinforcement learning approaches such as model-free and model-based learning and the tradeoff between exploration and exploitation.

Kursinhalte

• Introduction to Reinforcement Learning
  • Understanding Reinforcement Learning
  • Components of Reinforcement Learning Systems
• Markov Chains
  • Markov Decision Process & Markov Property
  • Value Functions and Discounted Value Functions
  • General Utility Function
  • Actions & Policy
  • Bellman’s Equation
  • Value Iteration
  • Markov Chain Monte Carlo (MCMC)
• Bandit
  • Single-Arm Bandit
  • Multi-Arm Bandit
• Q-Learning
  • Time-difference Learning
  • Reinforcement Learning with Neural Networks & Deep Q Learning
  • Experience Replay
  • Double Q-Learning
  • Delayed Sparse Rewards
  • Hierarchical Learning
  • Value- vs Policy-Based Learning
  • Actor Critic Learning
• Reinforcement Learning Approaches
  • Model-Free Learning
  • Model-Based Learning
  • Exploration vs Exploitation

Fakten zum Modul: Niveau: Master · Unterrichtssprache: English · 5 ECTS-Punkte · Modulprüfung: Hausarbeit

Building a successful data-based product requires a significant amount of high-quality code which needs to run in a professional production environment. This course starts by introducing the agile approaches Scrum and Kanban and then discusses the shift from more traditional software development approaches to the DevOps culture.

Special focus is given to the discussion and understanding of techniques and approaches for producing high-quality code such as unit and integration testing, test-driven development, pair programing, and continuous delivery and integration.

Since many software artefacts are accessed via APIs, this course introduces concepts of API design and paradigms.

Finally, this course addresses the challenges of bringing code into a production environment, building a scalable environment, and using cloud-cased approaches.

Kursinhalte

• Agile Project Management
  • Introduction to SCRUM
  • Introduction to Kanban
• DevOps
  • Traditional lifecycle management
  • Bringing development and operations together
  • Impact of team structure
  • Building a DevOps infrastructure
• Software Development
  • Unit & integration test, performance monitoring
  • Test-driven development & pair programing
  • Continuous delivery & integration
  • Overview of relevant tools
• API
  • API design
  • API paradigms
• From Model to Production
  • Building a scalable environment
  • Model versioning and persistence
  • Cloud-based approaches

Fakten zum Modul: Niveau: Master · Unterrichtssprache: English · 5 ECTS-Punkte · Modulprüfung: Mündliche Prüfung

Machine learning is a field of scientific study concerned with algorithmic techniques that enable machines to learn performance on a given task via the discovery of patterns or regularities in exemplary data. Consequently, its methods commonly draw upon a statistical basis in conjunction with the computational capabilities of modern computing hardware.

This course aims to acquaint the student with the main branches of machine learning and provide a thorough introduction to the most widely used approaches and methods in this field.

Kursinhalte

• Introduction to Machine Learning
  • Regression & Classification
  • Supervised & Unsupervised Learning
  • Reinforcement Learning
• Clustering
  • Introduction to clustering
  • K-Means
  • Expectation Maximization
  • DBScan
  • Hierarchical Clustering
• Regression
  • Linear & Non-linear Regression
  • Logistic Regression
  • Quantile Regression
  • Multivariate Regression
  • Lasso & Ridge Regression
• Support Vector Machines
  • Introduction to Support Vector Machines
  • SVM for Classification
  • SVM for Regression
• Decision Trees
  • Introduction to Decision Trees
  • Decision Trees for Classification
  • Decision Trees for Regression
• Genetic Algorithms
  • Introduction to Genetic Algorithms
  • Applications of Genetic Algorithms

Fakten zum Modul: Niveau: Master · Unterrichtssprache: English · 5 ECTS-Punkte · Modulprüfung: Examen, 90 Minuten