Storing Memories as Attractors
Demonstration of Hopfield Net
Get hopfield from CBN site

Example of Pattern Restoration
Example of Pattern Restoration
Example of Pattern Restoration
Example of Pattern Restoration
Example of Pattern Restoration
Example of Pattern Completion
Example of Pattern Completion
Example of Pattern Completion
Example of Pattern Completion
Example of Pattern Completion
Example of Association
Example of Association
Example of Association
Example of Association
Example of Association
Applications of
Hopfield Memory
Pattern restoration
Pattern completion
Pattern generalization
Pattern association

Hopfield Net for Optimization and for Associative Memory
For optimization:
we know the weights (couplings)
we want to know the minima (solutions)
For associative memory:
we know the minima (retrieval states)
we want to know the weights

HebbÕs Rule
ÒWhen an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth or metabolic change takes place in one or both cells such that AÕs efficiency, as one of the cells firing B, is increased.Ó
ÑDonald Hebb (The Organization of Behavior, 1949, p. 62)

Example of Hebbian Learning:
Pattern Imprinted
Example of Hebbian Learning:
Partial Pattern Reconstruction
Mathematical Model of Hebbian Learning for One Pattern
A Single Imprinted Pattern is a Stable State
Suppose W = xxT
Then h = Wx = xxTx = nx
since
Hence, if initial state is s = x, then new state is s¢ = sgn (n x) = x
May be other stable states (e.g., Ðx)

Questions
How big is the basin of attraction of the imprinted pattern?
How many patterns can be imprinted?
Are there unneeded spurious stable states?
These issues will be addressed in the context of multiple imprinted patterns

Imprinting Multiple Patterns
Let x1, x2, É, xp be patterns to be imprinted
Define the sum-of-outer-products matrix:

Definition of Covariance
Consider samples (x1, y1), (x2, y2), É, (xN, yN)

Weights & the Covariance Matrix
Sample pattern vectors: x1, x2, É, xp
Covariance of ith and jth components:

Characteristics
of Hopfield Memory
Distributed (ÒholographicÓ)
every pattern is stored in every location (weight)
Robust
correct retrieval in spite of noise or error in patterns
correct operation in spite of considerable weight damage or noise

Stability of Imprinted Memories
Suppose the state is one of the imprinted patterns xm
Then:

Interpretation of Inner Products
xk × xm = n if they are identical
highly correlated
xk × xm = Ðn if they are complementary
highly correlated (reversed)
xk × xm = 0 if they are orthogonal
largely uncorrelated
xk × xm measures the crosstalk between patterns k and m

Cosines and Inner products
Conditions for Stability
Sufficient Conditions for Instability (Case 1)
Sufficient Conditions for Instability (Case 2)
Sufficient Conditions for Stability