@jhawkins discusses a new idea for learning behavior models. It involved trajectory memory (a sequence of movements) instead of movements in a reference frame and could be modeled in the same column as morphology. We discuss questions and implications around this idea.
Jeff also discusses a new idea around shared learning, which involves broadcasting the input from the column at the fixation point to many other columns during learning.
Summary Video
Main Video
0:00 Introduction
1:42 Learning of Trajectories in L3
20:50 Does This Always Require Smooth Pursuit?
31:58 How This Could Inform Motor Output
34:25 Issues and Discussions Points
44:31 Behaviors can be associated with a Child Object
49:21 Where Directional Sensitive Cells are Found
1:01:24 Object Orientation and Behavior Orientation
1:11:00 How Could This Look Like on the Example of the Stapler?
1:36:25 Shared Learning Problem and Idea