@jhawkins presents several problems that arise when we start going from one to multiple columns. They include the shared learning problem, sharing pose during inference, columns moving on and off the object, and combining permanent models with temporary features. We discuss these issues and some ideas around them.
Summary Video
Main Video
0:00 Introduction
3:33 Problems Related to Multiple Columns Modeling The Same Object
28:28 Could Object ID Cause Anchoring of Grid Cells?
48:09 Could Attention Area Be Used to Determine Relative Pose?
1:12:57 Problems Related to Classes and Forking Models (When Do We Create A New Model?)
1:15:53 Questions and Observations