Sunday, May 17, 2015

Associative memory


In summary, the Hopfield networks are an intelligent self-content addressable system. It recovers bad input based on previous experience, introduced on the system on the training phase. Hopfield network is based on our capability to recognize objects even under noisy environment.


 
Associative memory seen by image reconstruction. 

On a microarrays experiment, thousand of genes are inferred simultaneously. This may be used for recording known gene based on experiments and use them for future systems. The system supposes to associate future images to memories inside the net. Each spot represents a gene.
The “trick” is that not just the gene by its own makes up its identity, rather the “partners”. On layman words, gene does not work alone or randomly activating neighbors; connection is  lucrative, however they represent cost for the cell. 

Associative memory inspired by gene expression seen by microarrays


Microarrays gives a matrix with colors, a “heat table”, see scheme. The experiment loosely speaking is: a) two samples of the organism is separated and exposed to different conditions; b) the mRNA of those is separated; c) it is done a “color attaching”; d) the sample is scanned by red laser and stored on a computer; e) the sample is scanned by green laser and stored on a computer; f) the result is overlapped. The result is a matrix with colors that indicates each gene that is activated on each situation, not activated or partially activated. 

A very nice and simple way to understand associative memory is picturing you seen a letter attached to a windown glass, it is raining, notwithstanding the water creating interference in the image, you still can see the "T", this is similar to some experiments people used to send by email, where an image used to give different impressions in different people. You see what you can see, what is inside you head. Our brain is always making pattern reconstruction, this is a normal task for us. See that besides it is applied to image herein, it can be easily extended to sound reconstruction, see that most of the hearing we do is reconstruction, it accelerate our talks. 

Attachment







No comments:

Post a Comment

Systems Biology and Machine Learning

Systems Biology and Machine Learning