Machine Learning
Machine Learning
Machine Learning
https://blog.resellerclub.com/wp-content/uploads/2017/08/machine-learning-blog-banner.j
pg
https://www.iuvotech.com/tony-stark-needs-it-dr-plan/
https://xkcd.com/1838/
What is Machine Learning?
https://xkcd.com/1838/
Common Applications
● Classifying documents (e.g., “is this email spam?”)
● Labeling/captioning images (e.g., “who’s in this picture?”, “what’s happening
in this picture?)
● Predicting the future: weather, finance, medical outcomes
● Recommending content (e.g., movies, books, music)
● Fraud detection (e.g. detecting money laundering)
● Self driving cars (a combination of a number of these and other problems)
Today
ML a normal part of our lives:
● Video/image processing
● Speech/language processing
● Controlling robots
● Computational biology
● Healthcare analysis
● Annoying recommendation systems
When people say “AI” these days, they usually mean “ML”
● Seeds (algorithms)
● Nutrients (data)
● A gardener (ML expert)
● The difference between predicted and actual, squared, summed across all
examples
● Find the line that minimizes squared error (in 2 dimensions, plane in 3
dimensions...)
○ Does this by choosing weights (m, b); changing the weights moves the line.
● How do we find the best weights?
○ Derivatives of the loss: see MATH 12X series
○ Tons of features (high dimensional) data?
○ Expensive/difficult to just compute the whole gradient
○ Loss functions weird in high dimensions
How does training work/Gradient Descent
Let’s say we’re playing a fun game: Blindfolded Mountain Descent!
1998 2014
1998 2010-
Neural Networks in Python
Unless you’re dealing with very high-dimensional data (images, audio, text), it’s
probably overkill. But it’s really powerful!