[Ml-stat-talks] tomorrow, yael niv, 1230

David Blei blei at CS.Princeton.EDU
Tue Apr 17 20:16:51 EDT 2012


hi ml-stat-talks

tomorrow: local talent.  yael niv will discuss "learning to
selectively attend" at 12:30pm in CS room 402.  come a bit earlier for
pizza.  the abstract is below.

best
dave

--

Yael Niv
CS402, 1230PM
Learning to selectively attend

Theories of learning often rely on learning the values of different
actions at different states, in terms of their potential for future
reward. However, in the real world "states" are extremely
multi-dimensional, such that even the simplest tasks are seemingly
impossible to learn due to the curse of dimensionality. Luckily, only
few aspects of our environment are actually relevant for each task, so
we can safely ignore (and generalize over) the rest. How do we learn
what dimensions to attend to in learning and what to ignore? This talk
will center around a simple multi-dimensional bandit task that we
study in my lab (and that Bob presented briefly last week), discussing
different possible models for learning in the task, and what human
behavior and neural activations tell us about these alternative
models. I will make the talk interactive: one of you will get to play
the task, and everyone will get to try their hand in modeling it!


More information about the Ml-stat-talks mailing list