Data Science Decisions in Time: Information Theory & Games

Overview

Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off

One annual plan covers every course and certificate on Coursera. 40% off for a limited time.

This is part of our specialization on Making Decision in Time. For this third course we start with an intriguing study on SFPark and build new insights into the ideas that flow from this direction. The ending point should bring new code and new algorithm insights into perspective, and use, by many computer and data scientists.

Syllabus

Dynamically Changing Environments and Game Theory

How should a control be adjusted to best achieve a desired outcome? We introduce the SFPark problem, a real parking management approach being used in SF. The question that we want to understand, via sequential methods and games, is how best to set the prices for spaces, dynamically during the day, to encourage a particular (say 15%) free space availability. The game is between the consumers (looking for parking) and the city (trying to optimize space, reducing those cruising for spaces and encouraging those coming for a meal or for shopping to have a parking space). This is a sequential decision problem that can also be described as a game.

Cooperative Games: How to Make Decisions with Missing and Ambiguous Information

Decision making as a shared endeavor rapidly extends game theory into many real world situations and helps us to see how these ideas can be applied to problems that impact all of us. We start with a discussion about water resources and their allocation. This then is tied back to how we think about the classic problem of the prisoner's dilemma.

Predicting and Understanding Your Game Opponents

For many real-world settings we are not fully cooperative and may even be playing a game with antagonistic opponents. Understanding an optimal strategy for these settings means paying attention to the moves possible from the opponent and what they mean for your own optimal actions. We start with considerations of cybersecurity and then move into the classic Centipede Game.

Sequential Social Environments: Optimal Play

The game of Diplomacy is a challenge due to the many combinatorial options that can flow from a set of decisions. The game can be quite complex to play and also provides an excellent training ground for computer algorithms. In this part of the course we look at the general nature of complex social interactions and the models for game play that can be used to define optimal policies.

Seeing Relevant Information - Avoiding Distribution Shifts and Being Relevant

In this fifth module we aim to generalize from our study of games as objects in their own right to algorithms and informational settings where the ideas from game theory can inspire new insights and ways to see into large and diverse datasets. We start with a common clinical problem: how to classify a radiological image. As we think about the challenges of this setting, including extracting and seeing the relevant features, we set the frame for our goals with this fifth week. In particular, how can we find the most important, and ideally invariant, features that best describe our problem and that can be used for making decisions.