Nov 22
aside
computer vision
Finish discussion of computer vision.
- discuss homework (Sam, at least)
Two examples:
Note particularly that people often use many steps in the processing,
including the learning techniques we've discussed earlier as
well as searching databases of known examples.
Other issues we haven't discussed: camera calibration, optics, stereo imaging, ... there's a lot of ongoing research in this area.
next topic : markov models
Today we're going to talk about the "big picture", *not* the math details. But we will look at some of those details - I've included some references here which I'd like you to look over; we will look into more detail next week.
Walk through
- basic idea of a "markov chain"
- generalization to a "hidden markov model"
- connections to finite state machines and formal grammers
The sorts of questions one would like to answer :
- Given an HMM model, generate a series of observations.
- Given a model and N observations, what is the most likely k-th state?
- What is the most likely sequence of of all N states?
- Given the observations, what is the most likely model?
... and there are a variety of well known math-ish (matrix multiplication and Baysian probability) algorithms to do these things; that's why these are popular models, even though
in practice they are somewhat simplistic.
A few examples of generators :
Readings
Textook:
- AIMA chap 15, "probabilistic reasoning over time"
- AIMA chap 22 & 23, on natural language processing
ai-class.com:
- video lecture in section 11, HMMs and Filters
wikipedia:
other:
