Upgrade to Pro — share decks privately, control downloads, hide ads and more …

The Learning Machine

The Learning Machine

Artificial Intelligence, in the form of Machine Learning (ML), has already transformed medicine, retail sales, and other industries. It is about to enter all our lives, whether we want it or not! In this session, we will quickly develop a basic understanding of AI, and its advantages, applications, and difficulties.

Interested in content like this? Then check out The Free-Range Technologist from Prof C. https://frtech.substack.com/

J. Scott Christianson

September 15, 2020
Tweet

More Decks by J. Scott Christianson

Other Decks in Education

Transcript

  1. Meet the Learning Machine
    J Scott Christianson
    Associate Teaching Professor
    How Artificial Intelligence is transforming our world!

    View full-size slide

  2. Artificial Intelligence (AI)
    Definitions
    Artificial Intelligence: A machine that exhibits
    cognitive or decision-making behavior and
    can take action to achieve a goal.

    View full-size slide

  3. • General AI: A machine that can
    reason and adapt like a human. E.g,
    sci fi movies.
    Artificial Intelligence (AI)
    Definitions
    Artificial Intelligence: A machine that exhibits
    cognitive or decision-making behavior and
    can take action to achieve a goal.

    View full-size slide

  4. • General AI: A machine that can
    reason and adapt like a human. E.g,
    sci fi movies.
    Artificial Intelligence (AI)
    Definitions
    • : A machine that is
    optimized for a particular task or
    project.
    Narrow AI
    Artificial Intelligence: A machine that exhibits
    cognitive or decision-making behavior and
    can take action to achieve a goal.

    View full-size slide

  5. Machine Learning and Deep Learning
    Narrow AI
    Machine Learning
    Deep Learning
    Machine Learning’s goal is to
    develop predictions based on
    previously observed patterns.
    Various variables are weighed to
    predict the probabilities of the
    outcomes. The variables and
    formula used to make such
    predictions may be programmed
    by a human OR
    they can be developed by the
    machine itself (Deep Learning).

    View full-size slide

  6. Deep Learning
    01
    Collect
    Training
    Data
    02
    Analyze
    and
    Segment
    03
    Setup and
    Train a
    Neural
    Network
    04
    Test and
    deploy

    View full-size slide

  7. solid
    vertical
    diagonal
    horizontal
    e2eml.school
    Deep Learning
    Input
    Case 1
    Case 2
    Case 3
    Case 4

    View full-size slide

  8. solid
    vertical
    diagonal
    horizontal
    e2eml.school
    Deep Learning
    Input
    Case 1
    Case 2
    Case 3
    Case 4

    View full-size slide

  9. solid
    vertical
    diagonal
    horizontal
    e2eml.school
    Deep Learning
    Input
    Case 1
    Case 2
    Case 3
    Case 4

    View full-size slide

  10. Types of Data For ML Processing
    • Motion Data

    • Audio/Voice Data

    • Image Data

    • Text Data

    • Geospatial Data

    • Physiological/Medical
    Data

    • Financial Data

    • Behavioral Data

    • and Much More

    View full-size slide

  11. Types of Data For ML Processing
    • Motion Data

    • Audio/Voice Data

    • Image Data
    • Text Data

    • Geospatial Data

    • Physiological/Medical
    Data

    • Financial Data

    • Behavioral Data

    • and Much More

    View full-size slide

  12. Figure 2: Applications of AI algorithms in medicine. The left panel shows the image fed into an
    algorithm. The right panel shows a region of potentially dangerous cells, as identified by an
    algorithm, that a physician should look at more closely. (From Artificial Intelligence in Medicine:
    Applications, implications, and limitations by Daniel Greenfield.)

    View full-size slide

  13. More than 50 AI/ML algorithms have been cleared by the
    US Food and Drug Administration for uses that include
    identifying intracranial hemorrhage from brain computed
    tomographic scans and detecting seizures in real time.
    Algorithms are also used to inform clinical operations, such
    as predicting which patients will “no show” for scheduled
    appointments. More recently, algorithms that predict in-
    hospital mortality have been proposed to inform ventilator
    allocation during the coronavirus disease 2019 pandemic.

    JAMA Article by Stephanie Eaneff, MSP1,2; Ziad Obermeyer, MD3; Atul J. Butte,
    MD, PhD2,4

    View full-size slide

  14. Types of Data For ML Processing
    • Motion Data

    • Audio/Voice Data
    • Image Data

    • Text Data

    • Geospatial Data

    • Physiological/Medical
    Data

    • Financial Data

    • Behavioral Data

    • and Much More

    View full-size slide

  15. Interactive Voice Response
    IF..Then Based Systems

    View full-size slide

  16. Interactive Voice Response
    ML Based Systems

    View full-size slide

  17. When and How to Use ML
    • Autonomous Vehicles
    AI and Ethics

    View full-size slide

  18. When and How to Use ML
    • Autonomous Vehicles

    • Admissions and Grading
    AI and Ethics

    View full-size slide

  19. When and How to Use ML
    • Autonomous Vehicles

    • Admissions and Grading

    • Loans and Credit
    AI and Ethics

    View full-size slide

  20. When and How to Use ML
    • Autonomous Vehicles

    • Admissions and Grading

    • Loans and Credit

    • Social Media
    AI and Ethics

    View full-size slide

  21. When and How to Use ML
    • Autonomous Vehicles

    • Admissions and Grading

    • Loans and Credit

    • Social Media

    • Warfare
    AI and Ethics

    View full-size slide

  22. Problems with AI

    View full-size slide

  23. solid
    vertic
    diagonal
    horizontal
    e2eml.school
    Input
    Case 1
    Case 2
    Case 3
    Case 4
    Problems with AI
    “Hidden Layers”

    View full-size slide

  24. Problems with AI
    Adversarial AI
    from Savan Visalpara

    View full-size slide

  25. Problems with AI
    Adversarial AI
    from Weijia Zhang

    View full-size slide

  26. Problems with AI
    Adversarial AI
    from MIT CSAIL

    View full-size slide

  27. Problems with AI
    Adversarial AI
    Images: Evtimov et al
    Camouflage graffiti and art stickers cause a neural network to
    misclassify stop signs as speed limit 45 signs or yield signs.

    View full-size slide

  28. http://LearnAbout.AI

    View full-size slide