From stock investment to autonomous vehicles: Artificial intelligence takes the world by storm. In many industries such as healthcare, transportation or finance, smart algorithms have become an everyday reality. To be successful now and in the future, companies need skilled professionals to understand and apply the powerful tools offered by AI. This course will help you to achieve that goal.
This practical guide offers a comprehensive overview of the most relevant AI tools for reasoning under uncertainty. We will take a hands-on approach interlaced with many examples, putting emphasis on easy understanding rather than on mathematical formalities.
After this course, you will be able to...
… understand different types of probabilities
… use Bayes’ Rule as a problem-solving tool
… leverage Python to directly apply the theories to practical problems
… construct Bayesian networks to model complex decision problems
… use Bayesian networks to perform inference and reasoning
Wether you are an executive looking for a thorough overview of the subject, a professional interested in refreshing your knowledge or a student planning on a career into the field of AI, this course will help you to achieve your goals.
What am I going to get from this course?
- The opportunity to understand and explore one of the most exciting advances in AI in the last decades
- A set of tools to model and process uncertain knowledge about an environment and act on it
- A deep-dive into probabilities, Bayesian networks and inference
- Many hands-on examples, including Python code
- A firm foundation to further expand your knowledge in AI
Module 1: Welcome to this course!
Introduction & Course Overview
In this lecture, I will introduce you to the course, its main goals and topics as well as its significance in the field of AI. Also, I will briefly introduce myself as your instructor and mentor on this journey.
Introductory Example: Medical Diagnosis
In this lecture, we will look at an introductory example from the field of medical diagnosis. The goal is to develop a feel for probabilities and for the deceptive properties of human intuition.
Module 2: Quantifying Uncertainty - Probability Theory and Bayes' Rule
In this lecture, you will learn about the various types of agents in AI and the differences between them. Also, I will introduce the agent type we will be concerned with in this course.
How to deal with uncertainty
In this lecture, you will learn about the major approaches with which to address uncertainty. You will learn about logic, sentences and models.
Example "Predicting a Burglary" (logic-based)
In this first example, we will try to predict wether our alarm has been triggered by an earthquake or by an actual burglary.
Basic probability theory
In this lecture, you will learn that probabilities are an effective way of dealing with gaps in models or in data we observe. Also, I will introduce random variables as a means to build a model of an environment.
In this lecture, we look at various types of probability and the differences between them. We will focus on conditional probabilities, which are a prerequisite for understanding Bayesian concepts.
In this lecture, I will introduce Bayes' Rule, one of the cornerstones of modern AI. You will learn how this simple rule allows us to reverse the order between what we observe and what we want to know.
Example: "Pedestrian Detection Sensor"
In this example, the reliability of a sensor for detecting pedestrians is assessed using Bayes' Rule. Relying only on its sensors, an autonomous vehicle has to decide wether to issue an emergency breaking or not.
Example "Clinical Trial" (with Python code)
In this example, we will apply Bayes' Rule to a scenario surrounding a clinical trial. In addition to solving some equations on our own, we will also make use of Python to facilitate computation.
Module 3: Representing uncertainty - Bayesian networks and probability models
In this lecture, you will learn how evidence from multiple sources can be combined to formulate more complex queries. Also, you will learn about the Naive Bayes Model, a concept in AI that works surprisingly well in practice.
Constructing Bayesian Networks
In this lecture, I will introduce Bayesian networks as a tool to graphically model relationships between multiple conditionally independent random variables.
Reasoning in a 2-node Network
In this lecture, we will focus on how to update the belief into a random variable by using the law of total probability and Bayes' rule. Also, we will look at how inference is performed in this simple setup.
Reasoning in a 3-node Network
In this lecture, I will introduce causal, diagnostic and inter-causal inference. Depending on the available evidence and on the direction of reasoning within the network, we will look at how inference is performed in this slightly more complex setup.
Polytrees and Belief Propagation
In this lecture, we will look at networks where there is at most one path between any pair of nodes. Also, you will learn about a standard algorithm for performing inference called 'belief propagation'.
Example "Predicting a Burglary" (extended)
In this example, we will expand the burglary scenario by adding more variables and modeling them into a Bayesian network. We will also illustrate the workflow of the message passing algorithm.
Example "Predicting a Burglary" (in Python)
In this example, I will introduce the Python toolbox 'pgmpy' as a mighty software to model Bayesian networks and answer queries using inference algorithms such as message passing. Finally, I will show how to take decisions based on probability distributions within the network.
Module 4: Summary & final remarks