Welcome to the final lecture on Predict.
This week, we'll start with bottom up
prediction.
How we can predict future values of data
from the data that we have.
We've seen some of this already in
learning.
How we can predict what future instances
look like or are part of based on past
experience.
But we'll go on to deal with predicting
values rather than simply classes.
We'll begin with the very basic prediction
technique, which is least-squares
approximation, and function approximation,
in general.
Then, we'll go on to show the relationship
between prediction, optimization, and then
controlling what you want to do with your
predictions.
Next, we'll ask how the brain actually
does all these things in a very smooth and
almost invisible manner and talk about
some recent developments in something
called hierarchical temporal memory, which
is a prediction system modeled after how
the brains and neurons actually are put
together.
And it is the latest advance in what was
originally the field of neural networks
but we'll see that neural networks, belief
networks and everything is sort of coming
together in this fairly interesting
development.
However, it turns out that bottom up
prediction so far, can go only a certain
way and one needs to combine many
different techniques including symbolic
reasoning as well as direct learning from
the data.
And we will discuss a very popular and
fairly old architecture called the
blackboard architecture which is becoming
more and more important, as all these
techniques start working together in large
complex systems.
Of course then, we'll finally summarize
seeing where we've come in this course
about what we've learned about web
intelligence, the brain, and adaptive
business intelligence based on all these
techniques.
And lastly, I'll leave you with some
challenge problems, which are fairly deep
and might interest some of you to actually
take these up for research.