Title: Back to Basics: Classification and Inference Based on Input Feedback Structure
1Back to Basics Classification and Inference
Based on Input Feedback Structure
Department of Computer Science University of
Illinois at Urbana-Champaign
2AI -gt AGI
- Ability to generalize
- Even if only learned basics
- Training distribution ? test distribution
- Avoid Combinatorial Explosion
- Allows complex networks
3New Basic Computational Structure
- Based on massive feedback to inputs
- No emphasis on weight parameters
- Input Feedback during testing
4Avoids Combinatorial Explosionvia Simple
Connectivity
Y1
Y3
Y4
Output Nodes
I4
Input Nodes
x4
x1
x2
x3
Connections
Positive
Negative
5Iterative
Y2
Y1
I2
I1
x1
x2
1
thus Wy
6Back
Y2
Y1
I2
I1
x1
x2
7Forward
Y2
Y1
I2
I1
x1
x2
8Back
Y2
Y1
I2
I1
x1
x2
9Active (1)
Y2
Y1
Inactive (0)
I2
I1
10Active (1)
C2
Inactive (0)
I2
I1
11Active (1)
C2
Inactive (0)
I2
12Active (1)
Inactive (0)
13Active (1)
Inactive (0)
14Active (1)
Inactive (0)
I2
15Active (1)
Inactive (0)
16Steady State
Graph of Dynamics
1
Y1
Y2
Activity
0
I2
0
1
3
2
4
5
I1
Simulation Time (T)
17Resolving Pattern Interactions
Network Configuration
Steady State
Results Node?Value
1
2
Inputs
Node
Input
A
B
(0, ½)
A1 and B ½ (
1,
Half Activation
Half Response
½)
A.12 C1.12 A2.5 B1 c11.5 c21
18Resolving Pattern Interactions
Based on Available Representations
2
3
Cells
Inputs
A
B
C
19Resolving Pattern Interactions
Binding
Most efficient configuration
Not possible with OvA or AvA
20Can be Chained Ad Infinitum
2
3
N
Nodes
...
Inputs
N
O
A
B
C
N
1
2
3
Nodes
...
Inputs
N
O
A
B
C
21New data Recognize Scene When Trained on
Individuals
- Teach single letters
- Test multiple simultaneous letters
- A scene is beyond the training distribution
22Feature Extraction
512 features
- Found in Visual Cortex
- Pixels Separated into features
Inputs to Model
In
I3
Feature Examples
Feature 1 Feature 2 Feature 3 Feature 4 Feature
n
Feature 1
Feature 2
Feature 3
?
D X
Fig 4 Simple feature extractor presenting
non-spatial information from visual field
collective pixel patterns presented to
network
23Two Stimuli Simultaneously
100
90
80
70
60
of combinations
50
40
30
20
10
0
0/2
1/2
2/2
Letters Correctly Classified
Figure 5 NN with two letter retraining
24Four Stimuli Simultaneously
A B C D
100
90
80
70
60
50
of combinations
40
30
20
10
0
0/4
1/4
2/4
3/4
4/4
Letters Correctly Classified
Figure 6 Four Letter Classification
Figure 5 NN with two letter retraining
25Difficulty
- Nonlinear Equations
- Cant mathematically prove general properties
26Steps Towards AGI
- Generalize Outside Training Distribution
- Structure Avoids Combinatorial Explosion
27Acknowledgements
National Geospatial-Intelligence Agency
HM1582-06--BAA-0001
28Equations
Activation
Feedback
Inhibition
Combined
C collection of all output cells Ca cell
a. Na the set of input connections to cell
Ca. na the number of processes in set Na of cell
Ca. P primary inputs (not affected by shunting
inhibition). I collection of all inputs Ib input
cell b. Mb the set of recurrent feedback
connections to input Ib. mb the number of
connections in set Mb
Q shunting inhibition. Qb shunting inhibition at
input b.