INF2D: Week 9: Approximate Inference Methods, and Time
This week we will continue to study how an agent estimates the likelihood of factors in the domain that it can't see, given the things it can see.
So far, we have studied Bayesian Networks, and exact inference methods over these representations of domains. There are two outstanding issues, however. First, these exact inference methods are intractable in the worst case. Second, Bayesian Networks model static environments (ie., while each random variable has a range of possible values, its actual value doesn't change over time). We need the means to model uncertainty in dyanmic environments as well: ie, to model inference when the actual value of each random variable can change over time.
This week we will address both of these topics. The first section this week presents approximate inference methods for Bayesian Networks:
25: Approximate Inference in Bayesian Networks
The next two sections then introduce extensions to the basic ideas behind Bayesian Networks so as to represent and perform inference with uncertainty in dynamic environments:
26: Time and Uncertainty I
27: Time and Uncertainty II
As always, these lectures will be delivered in person and live streamed. But in addition, each of the above includes pre-recorded videos of the lectures (with edited captions), the slides that were used in the videos, required readings, and a post-lecture quiz. The quiz is a chance for you to gauge your understanding of the material presented here, and so we strongly encourage you to review this content in the above order, and then complete the quiz. If there is anything you don't understand, then you have several options:
- Post a question on piazza;
- Ask a question at the in-persion lecture; and/or
- Ask your tutor.
For coursework, you can also get your queries addressed by attending the demonstrators' teaching hour.