1 of 20

Artificial Intelligence and Machine Learning

2 of 20

Agenda

  • What is AI?
  • Modern Approaches
  • Resources

2

3 of 20

1.What is AI?

It’s not like the movies

4 of 20

What is AI?

  • Artificial Intelligence means many things to different people
  • Artificial General Intelligence (AGI)
    • Duplicating human intelligence
    • Most research is focused on other forms of AI
    • Popular media (movies, books) emphasizes AGI
    • “Make flying machines so much like pigeons that they can fool other pigeons” Russell and Norvig
  • If not AGI, then what is AI?

4

5 of 20

AI is Studying Rationality

  • Rationality
    • Doing the “right thing”
    • Based on measurable metrics and observations
  • Computational Agents
    • Act rationally
    • Operate autonomously
    • Perceive their environment
    • Persist over a prolonged time frame
    • Adapt to change
    • Create and pursue goals

5

6 of 20

Computational Agents

  • Significant research is devoted to computational agents
  • There is a wide range of what can be considered a computational agent
    • Autocomplete field
    • Chat bot for company help system
    • Recommendation system for songs, storefronts, Netflix
    • Fraud detection for billing systems
    • List of next items to show in social networking feed
    • Player for online game site

6

7 of 20

Where is AI Used?

  • Can the problem domain be only partially modelled?
  • Is there a lot of available data?
  • Is there a question with a finite range to be answered?
    • Yes/no questions
    • What letter is this?
    • What are the objects in the picture?
  • Is the environment very complex?
    • How does a teenager learn mathematics?
  • Is the environment only partially observable?
    • Autonomous car in a snowstorm on a partially covered road
  • Is the domain too large for search?
    • What is the best move for this go/chess agent?

7

8 of 20

2.Modern Approaches

A lot has happened in AI since it started

9 of 20

Knowledge Representation

  • Determine how to represent semantics
  • Combines logic with inference
  • Creates planning models

9

10 of 20

Uncertainty

  • Probability required
    • What is likely to be there?
    • Bayes’ Theorem
      • Next step on conditional probability
  • Combine with knowledge representation
    • What’s missing?
    • Bayesian Networks
  • Changes with time

10

11 of 20

Machine Learning

  • Learning from examples
  • Supervised learning
    • Tagging data
    • Provides expected output given input data
  • Unsupervised learning
    • Untagged data
  • Semi-supervised learning
    • Smaller tagged datasets
    • Can be combined with GANs and reinforcement learning
  • Deep learning
    • Can be supervised, unsupervised, or semi-supervised
  • Reinforcement Learning

11

12 of 20

Reinforcement Learning

12

13 of 20

Neural Networks

  • You’ve probably heard about Neural Networks
    • Based on biological neurons in the brain
    • The brain is still not fully understood, but it is evidence of natural intelligence
  • Artificial Neural Networks (ANN)
    • Mimic guesses of how neurons can interact with each other
    • First proposed in 1943 and developed since as mechanisms for how machines can “learn” based on external inputs
    • Backpropagation developed from the 1960’s to 1980’s to allow a network to modify itself
    • Deep learning developed in 2006 to improve high level concept recognition
    • Many different subclasses of ANNs

13

14 of 20

Alternatives to Neural Networks

14

15 of 20

3.Resources

So much to read

16 of 20

University Programs

  • Undergrad
    • Typically 3 - 5 courses starting in 3rd year
  • Graduate programs
    • Typically 5 - 10 courses
    • Wide variety of topics
  • MIT Opencourseware 6.034 Artificial Intelligence (2010)
  • MIT Opencourseware 6.S191 Deep Learning (2020)
  • UW CS 480 Machine Learning (Fall 2020)
  • Harvard CS50 AI with Python (edX)
  • UBC CS 322 Introduction to Artificial Intelligence (2020)

16

17 of 20

Online Courses

  • Coursera
  • Fast.ai
    • Lessons to make AI very accessible
    • Less data, better algorithms
    • Free and/or accessible libraries
  • Vector Institute
    • Chief Scientific Advisor Geoffrey Hinton (co-inventor of back propagation)
    • Research Institute from University of Toronto

17

18 of 20

Books

  • Artificial Intelligence A Modern Approach 4E (2020)
    • Standard reference for most introductory AI courses
  • Artificial Intelligence 2E Foundations of Computational Agents (2017)
    • Freely available online
  • Deep Learning (2016)
    • Freely available online
    • Harder to learn from compared with first two resources
  • Reinforcement Learning 2E (2018)
    • Freely available online
    • Great reference, highly mathematical
  • Pattern Recognition and Machine Learning (2006)
    • Still considered a necessary reference for ML
  • Elements of Statistical Learning (2009)
    • Freely available online

18

19 of 20

Additional Resources

19

20 of 20

Credits

Special thanks to all the people who made and released these awesome resources for free:

  • Presentation template by SlidesCarnival
  • https://commons.wikimedia.org/wiki/File:Neural_network_example.svg#/media/File:Neural_network_example.svg

20