Bioinspired Associative Memories

Bioinspired Associative Memories

Roberto A. Vazquez (University of Valladolid, Spain) and Humberto Sossa (Center for Computing Research, IPN, Mexico)
Copyright: © 2009 |Pages: 8
DOI: 10.4018/978-1-59904-849-9.ch039
OnDemand PDF Download:
No Current Special Offers


An associative memory AM is a special kind of neural network that allows recalling one output pattern given an input pattern as a key that might be altered by some kind of noise (additive, subtractive or mixed). Most of these models have several constraints that limit their applicability in complex problems such as face recognition (FR) and 3D object recognition (3DOR). Despite of the power of these approaches, they cannot reach their full power without applying new mechanisms based on current and future study of biological neural networks. In this direction, we would like to present a brief summary concerning a new associative model based on some neurobiological aspects of human brain. In addition, we would like to describe how this dynamic associative memory (DAM), combined with some aspects of infant vision system, could be applied to solve some of the most important problems of pattern recognition: FR and 3DOR.
Chapter Preview


Humans possess several capabilities such as learning, recognition and memorization. In the last 60 years, scientists of different communities have been trying to implement these capabilities into a computer. Along these years, several approaches have emerged, one common example are neural networks (McCulloch & Pitts, 1943) (Hebb, 1949) (Rosenblatt, 1958). Since the rebirth of neural networks, several models inspired in the neurobiological process have emerged. Among these models, perhaps the most popular is the feed-forward multilayer perceptron trained with the back-propagation algorithm (Rumelhart & McClelland, 1986). Other neural models are associative memories, for example (Anderson, 1972) (Hopfield, 1982) (Sussner, 2003) (Sossa, Barron & Vazquez, 2004). On the other hand, the brain is not a huge fixed neural network as had been previously thought, but a dynamic, changing neural network. In this direction, several models have emerged for example (Grossberg, 1967) (Hopfield, 1982).

In most of these classical neural networks approaches, synapses are only adjusted during the training phase. After this phase, synapses are no longer adjusted. Modern brain theory uses continuous-time model based on current study of biological neural networks (Hecht-Nielse, 2003). In this direction, the next section described a new dynamic model based on some aspects of biological neural networks.

Key Terms in this Chapter

Low-Pass Filter: Filter which removes high frequencies from an image or signal. This type of filters is used to simulate the infant vision system at early stages. Examples of these filters are the average filter or the median filter.

PCA: Principal component analysis is a technique used to reduce multidimensional data sets to lower dimensions for analysis. PCA involves the computation of the eigenvalue decomposition of a data set, usually after mean centering the data for each attribute.

Dynamic Associative Memory: A special type of associative memory composed by dynamical synapses. This memory adjusts the values of their synapses during recalling phase in response to input stimuli.

Stimulating Points: Characteristic points of an object in an image used during learning and recognition, which capture the attention of a child. These stimulating points are used to train the dynamic associative memory.

Random Selection: Selection of one or more components of a vector at randomly manner. Random selection techniques are used to reduce multidimensional data sets to lower dimensions for analysis.

Dynamical Synapses: Synapses that modified their values in response to an input stimulus also during recalling phases.

Associative Memory: Mathematical device specially designed to recall output patterns from input patterns that might be altered by noise.

Complete Chapter List

Search this Book: