search

brohrer / robot-brain-project

a general purpose learning agent
205 stars 46 forks source link

Meta parameters for attention and goal selection #7

Closed brohrer closed 9 years ago

brohrer commented 9 years ago

In a mechanism mimicking human emotion, Implement meta parameters for attention and goal selection.

These include arousal or fight/flight tendency to look ahead vs act immediately willingness to explore vs play it safe sadness? happiness? (recent reward history) anger/fear/anxiety? tendency to act quickly and (big punishment is expected) anticipation/excitement? tendency to strive (big reward is expected)

brohrer commented 9 years ago

I removed this issue from the 0.6.0 release

BECCA's enhancement strategy is very pragmatic: only add features when they can be justified by performance increases in a world. Another way to say it is that if I want to add feature X, I first need to devise a world in which that feature will significantly improve performance.

In order to justify meta-parameters or 'arousal', I need to create a world where the time scale for achieving rewards and avoiding punishments change changes dramatically over time. For instance, consider a squirrel gathering nuts that must occasionally avoid a hungry wolf. I'll include this in a later release.

brohrer commented 9 years ago

I've embarked on a deep reworking of BECCA to enable it to handle more complex worlds. It will also be a more plausible model of human brain function, as a fringe benefit. The new development effort obviates this task.