In this meeting @vclay summarizes our new ideas around rapid inference, object distortions, and affordances. The team discusses replay between sensations, shared attention across columns, and ways to recognize distorted objects. It also covers how affordances may be represented in the neocortex and support generalization.
Summary Video
Main Video
0:00 Introduction
0:44 Overview of the Brainstorming Week
5:51 Sharing Location Information for Flash Inference
6:06 What we Currently Do in Monty
15:16 What Does the Correct Solution Likely Look Like?
17:03 One Proposed Solution: Rapid Replay Between Sensations
31:57 All Columns Attending the Same Thing at the Same Time
37:30 Further Questions & Discussion
40:06 Does This Remove the Need of Communicating Locations?
45:28 Object Distortions: A Partial Solution plus Potential Additional Options
45:42 Recognizing Relative Feature Arrangement Even if 3D Morphology Changed
47:20 Solution: Use Movement in Lower Dimensional Space
1:23:53 When Does This Not Work?
1:26:32 Potential Additional Solutions: 1. Leverage Scale
1:35:34 2: Learn Separate Morphology Models for Different Distortions
1:39:04 3: Learn Models of how an Object Distorts and Apply Movements to M Model
1:39:52 4: Abstract/relative (but not metric) Arrangement of Features
1:48:29 5: Use Compositional Models with More Tolerance for Location Mismatches
1:48:58 Further Discussions
1:52:46 Affordances
1:53:27 What are Affordances (in Psychology)?
1:54:44 How Could Affordances be Represented in the Neocortex?
1:56:56 Generalize Affordances
2:03:46 Affordances in the Neocortex
2:11:41 Advantages of Affordances
2:12:21 Further Discussion