Recognizing locations on objects
Marcus Lewis — November 2017 [email protected]
Agenda
• The problem, in context: Detecting locations relative to objects
• Inspiration:Principles of grid cells
• A solution:Recall, Normalize, Vote
Part 1 The problem, in context: Detecting locations relative to objects
Consider the inputs to the brain.
Sensors move.The brain is very good at making sense of sensorimotor sequences.
Sensors are often independent.Familiar objects often cause novel combinations of sensory features.
The body is an array of sensors.With both vision and touch, the brain receives an array of sensory features.
?
Sensorimotor sequences
Model ofobjects
?
Features + Locations relative to object
solvable
Sensorimotor sequences
Model ofobjects
These locations
aren’t relative to the body.
These locations
aren’t relative to the body.
They’re relative to the object
They’re relative to the object
These locations
aren’t relative to the body.
?
Features + Locations relative to object
solvable
Sensorimotor sequences
Model ofobjects
solvable?
Features + Locations relative to body
solvable
This Talk
Sensorimotor sequences
Model ofobjects
Features + Locations relative to object
Given the sensor’s location relative to
the body…
…calculate its location relative
to the object.
Part 2 Inspiration: Principles of grid cells
A rat
An environment
Electrode
Entorhinal cortex
Cell
Electrode
Entorhinal cortex
Cell
Each grid cell fires at multiple equally spaced locations.
Electrode
Cell
Entorhinal cortex
A Room
Firing fields of 16 different cells.
Electrode
Cell 1 Cell 2Cell 3
1 2 3 4
8765
1211109
16151413
1 2 3
8765
1211109
16151413
1 2 3 4
8765
1211109
16151413
1
65
109
151413
1 2 3 4
8765
1 2 3 4
8765
1211109
161514
2 3 4
87
1211
16
1 2 3 4
8765
4 1
5
4
13 Entorhinal cortex
A Room
Room 1
After the rat learns a room, grid cells anchor to the
room consistently.
Room 2
They anchor to other rooms differently.
“Purely Translational Realignment in Grid Cell Firing Patterns Following Nonmetric Context Change”
Marozzi, Ginzberg, Alenda, Jeffery (2015)
A cool grid cell trick: Modules
Module 2Module 1
+ +
Module 3
Entorhinalcortex
= Big space of unique locations
Principles of grid cellsThe cell activity moves between a fixed
set of locations.
These locations are mapped onto the environment.
=+Body
Motion2. Updating it from motion cues
The brain determines the location by…
1. Recalling it from sensory cues +Feature
=
Part 3 A solution: Recall, Normalize, Vote
Sensory 1
Object
Input
Object
Input
Location Sensory 2Location
Column 1 Column 2
Layer 2
Layer 4
Column 1 Column 2
Object
Input
Feature relative to
Object
Object
Input
Feature relative to
Object
Body relative to
Object
Feature relative to
Body
Feature relative to
Body
New:Recall past locations of
sensed features.
New:Vote on the body’s location
relative to the object.
Layer 2
Layer 4
Layer 6A
Sensory 1 Sensory 2
BBABA
BBA
B
A A AB
AA B
AA
BA B
BA
A
A
Object 1
Object 3
Object 2
Object 4
BA B
BA
A
A
Demo:
BBABA
Visualizing the solution
All objects are learned at points in a large “location” space.
ABB
BBABA
All objects are learned at points in a large “location” space.
Visualizing the solution
BA
ABB
BBABA
All objects are learned at points in a large “location” space.
Visualizing the solution
BA
ABB
All objects are learned at points in a large “location” space.
BA B
BA
A
A
Visualizing the solution
BA B
BA
A
A
BA
ABB
All objects are learned at points in a large “location” space.
A
BA
Visualizing the solution
BA B
BA
A
A
BA
ABB
All objects are learned at points in a large “location” space.
AAB
BA
Visualizing the solution
BA B
BA
A
A
BA
ABB
All objects are learned at points in a large “location” space.
AAB
BBA
A
Visualizing the solution
BA
AA
AB
BBB
B
AA
This “location” space now contains maps of multiple objects.
Visualizing the solution
Map ofObject 1
Map ofObject 2
Distance between isn’t meaningful
BA
AA
AB
BBB
B
AA
BBABA
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
BA
AA
AB
BBB
B
AA
BBABA
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
BA
AA
AB
BBB
B
AA
BBABA
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
BA
AA
AB
BBB
B
AA
BBABA
2
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
BA
AA
AB
BBB
B
AA
BBABA
3 22
2 2
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
BA
AA
AB
BBB
B
AA
BBABA
3
To recognize a location, each sensor votes on the body’s location.
Visualizing the solution
solvable?
Features + Locations relative to body
solvable
This Talk
Sensorimotor sequences
Model ofobjects
Features + Locations relative to object
“Do I remember any objects that contain this
arrangement of features?”
The holistic system does this:
Summary
Location is probably one of the brain’s main primitives.
Grid cells suggest an approach to location: • Cortex has a “space” of locations that it can process. • Cortex maps objects into this space.
Cortex could recognize these locations in 3 steps: Recall, Normalize, Vote
BA
A
BB
Thank you.