data_mining:neural_network:hopfield [AE Wiki]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Hopfield nets ======

Architecture:
  * Recurrent connections
  * Binary theshold units

Idea:
If connections are **symmetric**, there is a **global energy function**.

Each binary configuration has an energy. Binary threshold decision rule causes network to settle to a **minimum of this energy function**.

===== Energy function =====
**Global energy:**

$E = - \sum_i s_i b_i - \sum_{i<j} s_i s_j w_ij$

  * $w_ij$ is connection weight.
  * Binary states of two neurons

Local computation for each unit:

$\Delta E_i = E(s_i=0) - E(s_i=1) = b_i \sum_j s_j w_{ij}$

===== Storing memories =====

 * memories could be energy minima of a neural net.
 * Binary threshold decision rule can the nbe used to clean up incomplete or corrupted memories.


  * Activities of 1 and -1 to store a binary state vector (by incrementing weight betweend any two units by the product of their activities). Bias is permantly on unit.
  * $\Delta w_{ij} = s_i s_j$

===== Storage capacity =====
N units = 0.15 N memories (At N bits per memory this is only 0.15N^2).

  * Net has $N^2$ weights and biases.
  * After storing $M$ memories, each connection weights has an integer value in the range of $[-M, M]$.
  * Number of bits required to store weights and biases: $N^2 log(2M+1)$.

===== Spurious minima limit capactiy =====
When new configuration is memorizes, we hope to create a new energy minimum (if two minima merge, capacity decreases).

==== Avoiding spurious minima by unlearning ====
Let net settle from random initial state, then do unlearning.

Better storage rule: Instead of trying to store vectors in one shot, cycle through training set many times. (Pseudo likelihood technique).

===== Hopfield nets with hidden units =====
Instead of memories, store interpretations of sensory input.

  * Input represented as visible units
  * Intepretation represented as hidden units.
  * Badness of interpreation rep. as energy.

Issues:
  * How to avoid getting trapped in poor local minima of the energy function? 
  * How to learn the weights on the connections to the hidden units and between the hidden units? 


===== Improve search with stochastic units =====

Hopfield net always reduces energy (trapped in local minima).
Random noise:
  * Lot of noise, easy to cross barriers
  * Slowly reduce noise so that system ends up in a deep minimum (Simulated annealing)

Stochastic binary units

* Replace binary threshold units with binary stochastic units that make biased random decisions ("temperature" controls noise amount; raising noise is equivalen to decreasing all the energy gaps betweend configurations)

$p(s_i=1) = \frac{1}{1+e^{-\Delta E_i/T}}$

===== Thermal equilibrium =====
Thermal equi. at temperature of 1

Reaching thermal equilibrium is difficult concept. Probability distribution over configurations settles down to statonary distribution.

Intuitively: Huge ensemble of systems that have same energy function. Probabiltiy of configuration is just fraction of the systems that have configuration.

Approaching equilibrium:

* Start with any distribution over all identical systems
* Apply stochastic update rule, to pick next configuration for each individual system
* May reach situation where fraction of systems in each configuration remains constant.
  * This stationary distribution is called thermal equilibrium.
  * Any given system keeps changing its configuration, but the fraction of systems in each configuration does not change.

===== Boltzman machine =====

Given: Training set of binary vectors. Fit model that will assign a probability to every possible binary vector.


Useful for deciding if other binary vectors come from some distribution (e.g. to detect unusual behavious).