Artificial Intelligence

Welcome to Introduction to Artificial Intelligence!

An introduction to basic AI concepts and the challenge of answering "what is AI?"

In this lesson, you'll dive right in and apply Artificial Intelligence to solve every Sudoku puzzle.

If you do not want to use Workspaces, then follow these instructions to set up your own system using Anaconda, a popular tool to manage your environments and packages in python.

Expand from the constraint propagation technique used in the Sudoku project to the Constraint Satisfaction Problem framework that can be used to solve a wide range of general problems.

Peter Norvig, co-author of _Artificial Intelligence: A Modern Approach_, explains a framework for search problems, and introduces uninformed & informed search strategies to solve them.

Peter introduces uninformed search strategies—which can only solve problems by generating successor states and distinguishing between goal and non-goal states.

Peter introduces informed search strategies, which means that they use problem-specific knowledge to find solutions more efficiently than an uninformed search.

Complete a practice exercise where you'll implement informed and uninformed search strategies for the game PacMan.

References to additional readings on search.

Peter Norvig returns to explain propositional logic and first-order logic, which provide a symbolic logic framework that enables AI agents to reason about their actions.

Peter Norvig defines automated planning problems in comparison to more general problem solving techniques to set the stage for classical planning algorithms in the next lesson.

Peter presents a survey of Classical Planning techniques: forward planning (progression search) & backward planning (regression search).

Peter discusses plan space search & situational calculus. Finish the lesson with readings on advanced planning topics & modern applications of automated planning.

Thad Starner introduces the concept of _iterative improvement problems_, a class of optimization problems that can be solved with global optimization or local search techniques covered in this lesson.

Thad introduces _Hill Climbing_, a very simple local search optimization technique that works well on many iterative improvement problems.

Thad explains _Simulated Annealing_, a classical global optimization technique for optimization.

Thad introduces another optimization technique: _Genetic Algorithms_, which uses a population of samples to make iterative improvements towards the goal.

Complete a classroom exercise implementing simulated annealing to solve the traveling salesman problem.

Review similarities of the techniques introduced in this lesson with links to readings on advanced optimization topics, then complete an optimization exercise in the classroom.

Extend classical search to adversarial domains, to build agents that make good decisions without any human intervention—such as the DeepMind AlphaGo agent.

Search in multi-agent domains, using the Minimax theorem to solve adversarial problems and build agents that make better decisions than humans.

Some of the limitations of minimax search and introduces optimizations & changes that make it practical in more complex domains.

Extensions to minimax search to support more than two players and non-deterministic domains.

Introduce Monte Carlo Tree Search, a highly-successful search technique in game domains, along with a reading list for other advanced adversarial search topics.

Welcome to Fundamentals of Probabilistic Graphical Models. In this lesson, we will cover the course overview, prerequisites, and do a brief introduction to probability.

Sebastian Thrun briefly reviews basic probability theory including discrete distributions, independence, joint probabilities, and conditional distributions to model uncertainty in the real world.

In this section, you'll learn how to build a spam email classifier using the naive Bayes algorithm.

Sebastian explains using Bayes Nets as a compact graphical model to encode probability distributions for efficient analysis.

Sebastian explains probabilistic inference using Bayes Nets, i.e. how to use evidence to calculate probabilities from the network.

Learn Hidden Markov Models, and apply them to part-of-speech tagging, a very popular problem in Natural Language Processing.

Thad explains the Dynamic Time Warping technique for working with time-series data.

Suggested resources to continue learning about artificial intelligence after completing the Nanodegree program.

Thad returns to discuss using Hidden Markov Models for pattern recognition with sequential data.

Thad shares advanced techniques that can improve performance of HMMs recognizing American Sign Language, and more complex HMM models for applications like speech synthesis.