Machine Learning on the High Trail - Introduction

Target group

This online course is aimed at all employees who want to know what is behind the term "machine learning". No prerequisites or mathematical knowledge needed.

Learning objectives

• Define Machine Learning
• Differentiate between the main types of machine learning systems
• Describe key challenges for machine learning projects
• Recognize why data-based methods represent a paradigm shift
  

Learning content

Introduction to Machine Learning by Examples

What is “machine learning”, how do machines “learn”, and what classes of problems can machine learning solve? We approach these questions by looking at two examples for „classification“ and „regression“. We will see that machines solve problems differently than humans, and that machines learn by example.

What is Machine Learning?

Machine learning algorithms devise statistical models by finding patterns in data in order to make predictions without being explicitly programmed. We consider several definitions of the term “machine learning“ and clarify the difference between two terms frequently used in the context of machine learning, “model” and “algorithm”. We get an overview of available machine learning methods by classifying them according to different criteria, such as the amount and type of supervision provided during training, or how the machine learning system generalises.

Key Challenges for Machine Learning

A machine learning model is expected to not just explain known data but predict future data. Challenges towards this task can be related both to the model and to the data. We discuss overfitting, a key challenge related to model selection that occurs when the model starts to fit features specific to the training data that do not generalize to new data; and sampling bias, an important data-related challenge the arises when the training data set is not representative of the new data the model is expected to generalize to.

Annotated Data Are a Scarce Resource

A crucial practical aspect for machine learning is the availability of sufficiently large training data sets. Especially large annotated data sets as required for supervised learning methods continue to be a scarce resource, and access to high-quality labeled data has become a major competitive advantage. In particular in production, obtaining annotated data can be very expensive, or even impossible. In these cases, other approaches are needed to train machine learning systems.

Machine Learning: Quo Vadis?

While traditional methods rely on pre-defined features selected based on human understanding, advanced machine learning algorithms autonomously learn relevant features from the data. With ever-increasing computational power and availability of data, the triumph of data-driven methods seems unstoppable.

As the power and sophistication of machine learning methods grow, so does our desire to explain why the machine yields a certain outcome or makes a specific prediction. We discuss an approach towards explainable machine learning systems and give a glimpse of the associated difficulties.

Test: Test your understanding of important concepts in machine learning.