Skip to content

Latest commit

 

History

History
74 lines (52 loc) · 6.08 KB

research.md

File metadata and controls

74 lines (52 loc) · 6.08 KB
layout title comments published
page
Research
false
false

Uncertainty often has a negative connotation. It implies some mismatch between how the world is operating and our ability to predict it. But for a learner, uncertainty is like a lighthouse pointing the way, signaling that there’s something worth learning about. Uncertainty can therefore be the link between action and learning.

My research centers on this interplay between uncertainty and action: how people make use of their uncertainty in order to decide how to learn about the world.

  • how is uncertainty represented by the mind?
  • how do people generate and test hypotheses about their environment?
  • how does active control of learning differ from passive conditions in which control is absent?
  • how do people balance the value of exploration against its costs?

See below for summaries of some ongoing projects, or the list of publications.

Uncertainty-driven spatial search

How does knowledge about possible states of the world guide the search for information?

battleshiptask{:class="img-float"} In this project, we examined exploration in a task similar to the board game Battleship. In this game, players were shown an empty grid and told that three shapes were hidden inside. These shapes were randomly sampled from a predefined set. By clicking on a square in the grid, they could learn whether a square in the grid belonged to a shape or was empty. Their goal was to simply learn the hidden configuration of shapes by uncovering the fewest squares as possible.

How do people transform their knowledge about the hypothesis space (i.e., different possible configurations of shapes from the original set) into decisions about which squares to uncover? We developed an “ideal searcher model” that quantifies the amount of uncertainty reduction expected from uncovering different locations (see example below). The model shows how uncertainty changes over the course of learning--a location that is highly useful on one trial may suddenly become redundant on the next.

battleshipmodel

{:refdef: .selectedpubs} Related publications:

Self-directed category learning

{:refdef: .selectedpubs} Selected publications:

Enhanced episodic memory through active control

{:refdef: .selectedpubs} Selected publications:

Adaptive exploration during experience-based decision making

{:refdef: .selectedpubs} Selected publications: