Neural Network Model Theories example essay topic

Theories of Knowledge and Psychological Applications Robin A. Finlayson University of Saskatchewan Ed. Psy: 855.3: Advanced Educational Psychology October 16, 1996 How individuals are able to obtain knowledge is something that psychologists have studied for a number of years. The ability to store and retrieve knowledge provides individuals with the propensity to form logical thought, express emotions and internalize the world around them. In order for a psychologist to understand the theories of knowledge it is necessary to investigate the aspects of the theories. In this paper we examine the history, the basic construct, the similarities of the theories and how those theories relate to psychological therapies. History of the theories The neural network model attempts to explain that which is known about the retention and retrieval of knowledge.

Neural network models have been examined for a number of years. In the mid 1940's and 1950's the first of the network models began to appear. These publications introduced the first models of neural networks as computing machines, the basic model of a self-organizing network (Arbib, 1995). In 1943 McCulloch and Pitts published their model theory (Arbib, 1995).

In 1948 Rashevsky proposed a number of neural network models to explain psychological phenomena. During this era not enough was known about the brain, subsequently he was considered ahead of his time. Rashevsky relied heavily upon complex mathematical equations within his model, consequently many people simply did not understand his theoretical perspective (Martindale, 1991). In 1958 Rosenblatt proposed his theory on neural network models which focused on perception. The theory elicited a great deal of interest; however it was considered too simple to sufficiently explain all aspects of perception (Arbib, 1995). As a result of the lack of acceptance, neural network models 'fell out of fashion' (Martindale, 1991, P. 12).

For a nine year lapse no neural network model theories were developed. In 1967 the network approach was again examined. Konorski developed a useful network model that focused primarily on Pavlovian conditioning as opposed to cognition. Grossberg developed his neural network theory during the years of 1969, 1980, 1987, and 1988. Grossberg developed a powerful network theory of the mind but, like the Rashevsky model, Grossberg's theory was comprised of complex mathematical terms and was therefore extremely difficult to understand.

His neural network models are only now being recognized as truly revolutionary (Martindale, 1991). Many new theorists would enter the field of neural network models, but it was the work of Rumelhart, Hinton, and McClelland that would simplify the way we would view such models (Arbib, 1995). It was in 1986 that Rumelhart, Hinton, and McClelland developed their network model. It was and still is regarded a sone of the most notable network theories.

This is true because they structured their theory in a clear, concise, and intelligible manner (Martindale, 1991). Neural network models have evolved during the past sixty years. The initial theories were extremely difficult to comprehend and they were not interchangeable with a broad range of topics. Today's theories are simpler to understand because they are less complex. The theories are capable of encompassing numerous topics. The dual coding approach is one that believes that knowledge is a series of complex associative networks.

Within these networks we find imaginal and verbal representations. These verbal and nonverbal representations are means that facilitate the retrieval and storage of knowledge (Paivio, 1986). The individual who was at the fore front of the development of the dual coding theory was Allan Paivio. He did research in the area of verbal and nonverbal representations during the 1960's. Research papers that dealt with topics of verbal and imaginal processes were: Abstractness, imagery, and meaningfulness in paired-associated learning (1965); Latency of verbal associations and imagery to noun stimuli as a function abstractness and generality (1966) and; Mental imagery in associative learning and memory (1969), (Paivio, 1986).

In 1971 Allan Paivio presented his revolutionary paper, Imagery and Verbal Processes. As a result of this paper the concept of a dual coding process was conceived. Paivio's subsequent paper in 1985, Mental Representations, retained the same constructive empiricism and the same basic theoretical assumptions as the earlier paper, Imagery and Verbal Processes. In this paper Paivio demonstrated that the fundamentals of a dual coding approach have stood up well to challenges over the years (Paivio, 1986).

The dual coding process offers a clear explanation of how individuals are able to store and retrieve knowledge. Through Paivio's dual coding approach we are able to see how internal networks of verbal and imaginal representations are capable of logging and retrieving information both non verbally and verbally. Construct of the theories There are a number of theories that explain how it is the human brain is capable of storing and retrieving information. A neural network model of cognition aims at explaining how and why we experience such mental phenomena.

The metaphor 'the mind works like a computer' has been heard by everyone at one time or another. Recently cognitive psychologists have considered that the mind does not work like a conventional computer. They have replaced the computer metaphor with a brain metaphor (Martindale, 1991). The logic for the rebuttal of the computer metaphor is that a computer has a central processing unit that is only capable of doing one thing at a time. It processes very quickly and in fact, operates at a million times faster than the average neuron (Arbib, 1995).

A computer can thus do long division problems quicker than you or I can, but there are some tasks-for example, perceiving and understanding a visual scene- that the brain can perform faster than a computer. In such a case, the brain could not possibly work like a computer. The brain therefore solves the problem of vision differently than a computer (Martindale, 1991). Martindlae (1991) states that 'The brain does not have anything we could really call a central processing unit, and the brain does not work in a serial fashion. The brain is therefor more like a large number of very slow computers all operating at the same time and each dedicated to a fairly specific task' (p. 10). Since the computer metaphor was replaced with the brain metaphor, a cognition model was needed to explain how and why we experience mental phenomena.

One such theory is the neural network model. A neural network model is composed of several components: 1. A set of possessing units, referred to as 'nodes' or 'cognitive units. '.

2. A state of activation. Nodes can be activated to varying degrees. These of these activated nodes corresponds to the contents of consciousness. The most active nodes represent what is being done at the time, all other deals with motor function at the unconscious level. 3.

A pattern of connections among nodes. Nodes are connected to one another by either or inhibitory connections that differ in strength. The strength of these connections constitutes long-term memory. 4.

Activation rules for the nodes. These rules specify such things as exactly how a node 'adds up " its inputs, how it combines inputs with its current state of activation, the rate at which its activation decays, and so on. 5. Output functions for the node. We assign thresholds or make output a nonlinear function of the node's activation, we get useful results. 6.

A learning rule. We need to explain how learning occurs; in a network model, learning means strengthening the connections between nodes. The connection between two nodes are strengthened if they are simultaneously activated 7. An environment for the system.

Neural network modules are massively interconnected. The nodes in any analyzer are organized into several layers. Connections among nodes on different layers are generally, and connections among nodes on the same layer are usually inhibitory. (Martindale, 1991). An interactive and competitive network consists of processing nodes gathered into a number of competitive pools. There are connections between pools and they are generally bidirectional.

Within the pool, the inhibitory connections are assumed to run from one node in the pool to all the other nodes in that pool, therefore they will not be activated (McClelland & Rumelhart, 1988). The easiest way to comprehend how a neural network model works is to examine a simple neural network model. Figure 1 is an interactive and competition model based on the works of McClelland (1991). The network model concerns knowledge about five people, this is represented by the five nodes in the center circle. There is nothing stored in these nodes. Knowledge about what they represent lie in their connections to the other nodes.

The attributes of the five Figure 1 (Martindale, 1991, p. 15) people are represented by nodes in the circles surrounding the center circle. Here is how the network works: The lines between circles indicate two way connections. We assume that the nodes within the circles have a inhibitory effect on one another. When any one node is activated it, inhibits nodes in its own circle and excites nodes to which it is connected in other circles.

These excited nodes go on to excite other nodes. Excitation and inhibition reverberates back and forth, some nodes will be activated and others will be inhibited. When one follows the lines back and forth we can see that the network stores information. For example Joe is a white male professor who drives a Subaru and likes brie cheese. It is also evident that Harold and Frank are both black stockbrokers, but one likes brie and the other likes cheese whiz (Martindale, 1991). The network has a number of properties that mimic the way people think.

First, all memory is content addressable. Stimulating the network with the word " Fred' activates the node that codes this name. Soon, the nodes coding these properties will be activated automatically. There is no need to search for information, simply saying the name 'Fred' automatically retrieves the information. Networks also show default assignments. The default assignment is the ability to hypothesize.

When the network is asked about Claudia, the node of brie cheese will be at least partially activated. This happens because the brien ode will receive activation from the node coding professors. This occurs because Claudia is a professor (Martindale, 1991). Although neural networks tend to become more complex than the example shown, it demonstrates why we experience mental phenomena. The network theory explains how we are able to retrieve information and then draw conclusions from that information. Another view or theory that attempts to explain mental phenomenon is the dual coding theory.

This theory uses verbal and nonverbal representations as the means by which individuals are able to store and retrieve information. Allan Paivio (1986) states: 'The theory is based on the general view that cognition consists of the activity of symbolic representational systems that are specialized for dealing with environmental information in a manner that serves functional or adaptive behavioral goals. This view implies that representational systems must incorporate perceptual, affective, and behavioral knowledge. Human cognition is unique in that it has become specialized for dealing simultaneously with language and with nonverbal objects and events. Moreover, the language system is peculiar in that it deals directly with linguistic input and output (in the form of speech or writing) while at the same time serving a symbolic function with respect to nonverbal objects, events, and behaviors. Any representational theory must accommodate this functional duality' (p. 53).

It is important to recognize that the general level of the dual coding theory divides into two subsystems, verbal and nonverbal. These two subsystems can be divided into sensorimotor subsystems, such as visual, auditory, hap tic, taste and smell (Paivio, 1986). When dealing with this theory it is important to remember that there is no top to bottom approach. This means that the activating mechanism can be either verbal or imaginal. For example the instruction to bring an image to words maximizes the probability that nonverbal representations will be activated by subsequent verbal cues (Paivio, 1986). When looking at verbal and imagery representations it is important to consider how they differ from one another.

The imagery or nonverbal system consists of a set of interconnected parts specialized for dealing with environmental information. The imagery system relies upon the nonverbal representations to provide feedback, these are visual, auditory, hap tic, taste, smell and other non linguistic representations. The verbal aspect utilizes words as codes. Objects, events or ideas can be encoded (Paivio, 1986). Another difference is how the two representations are organized. Paivio (1986) found that 'intra unit functional structures differ so that component information in higher-order nonverbal units are synchronously organized, where as verbal components are sequentially organized' (p. 59).

This means that imagery systems are able to evoke a number of representations at one time and are therefore capable of encoding much about a single complex image at one time. The verbal representation on the other hand must be made sequentially, only processing information one bit at a time. With a basic understanding for the inner workings of both the verbal and nonverbal representations it is important that we view the between- system relations. Although both systems would seem to be independent of one another, in that they are capable of being active without the other, it is evident that one system is capable of activating the other system. This would imply that if one system is capable of activating the other system they must be interconnected (Paivio, 1986). Although the two representational systems are capable of working independently they are also able to work together through interconnections.

This interconnection is known as a referential connection. The referential connection is the ability for one system (either verbal or nonverbal) to evoke the other and vise versa. Through this connection individuals are capable of describing and imagining any number of situations. Paivio (1986) states that 'the interconnections are not assumed to be one-to-one, but rather one-to-many, in both directions. The assumption parallels the familiar fact that a thing can be called by many names and a name has many specific references.

This translates into the dual coding assumption that a given word can evoke any number of images, corresponding to different exemplars of a referent class (e. g., different tables) or different versions of particular class member (e. g., my dinning room table imaged from different perspectives). Conversely, a given object (or imaged object) can evoke different descriptions' (p. 63). All that we hear, see, touch and smell is encoded into our verbal and nonverbal knowledge base. It is how we are able to store and retrieve these representations that make us capable of providing a verbal representation of an image in our minds, or enables us to imagine a verbal description. Comparisons and contrasts To have complete understanding of these two theories is important to compare and contrast them. It is important because commonalities allow for similar explanations of mental phenomena.

Both theories do an exceptional job of explaining the processes of the of the mind. One similarity between neural network theory and dual coding theory is that they both divide the components of their theory into subsets. The network theory puts the similar nodes into one set and the dual coding theory puts the verbal in one set and the imaginal into another set. Both theories utilize connections between subsets as a way of storing and retrieving knowledge.

While the theories have a number of similarities they also have some differences. The dual coding theory has two subsets, the verbal and the imaginal. The neural network theory has numerous amounts of nodes grouped into many different sets. These sets form webs. There are numerous webs layered one on top of the other and each is able to access one another. With the infinite number of webs being able to access one another the network theory has the potential to become more complicated than the dual coding theory.

Both theories make valid points as to how individuals process and retain knowledge. While the two theories may differ on the internal representations of the storage of knowledge, both have similar foundational beliefs: knowledge is taken in, it is stored, there are connections between the stored groups of knowledge and there is a retrieval process. How the theories apply to psychology Why is it important for a psychologist to know and understand the theories of knowledge? It is important because the field of psychology studies the processes of humans (how they act, react, develop, make decisions, cope, eat.

). If a psychologist has a basic understanding of the knowledge theories, then they will have a better understanding of the thought processes of a client. Therapies such as relaxation therapy, rational emotive therapy, art therapy and choice therapy must be able to appeal to the individuals knowledge constructs. Clients in cognitive therapy tend to posses irrational thoughts. In order to bring about change in the clients thought processes the therapists must assist the client to analyze their faulty logic. Through challenging what the client believes to be true the client is then able to analyze and reconstruct the knowledge that is stored in the verbal and imaginal compartments of the dual coding theory as well as the nodal compartments of the network theory.

In observing art therapy it is evident that the understanding of the knowledge theory would prove useful. Art therapy can be represented in three ways: it is experienced internally, it is expressed verbally, or constructed and represented through the media (Lusebrink, 1990). Lusebrink (1990) states that 'Internal experiences of images and there external representations influence each other... The internal image is based on sensory, affective, and thought processes. The image is externalized either through verbal descriptions or through the manipulation of media' (p. 6) In the above statement we can see a definite connection between art therapy and the knowledge theories. Through art therapy an individual must be able to view an image, internalize that image and be able to make the connection to express how that image expressed their feelings.

This is much the same as the knowledge theories. The theories of knowledge are tied directly to psychological therapies. The knowledge theories explain how a therapy technique is able to connect with a client's internal construct and assist in expressing or altering cognition. While absolute understanding of the knowledge theory may not be essential to an effective outcome of a therapy, it would assist in the understanding of how the therapy is able to work.

The theories of knowledge tend to be quite complex. In the terms of a psychological context it is important to understand the knowledge theories. The history, the construct, and their similarities all allow the psychologist to better understand how an individual internalizes the world around them. The basic understanding of the knowledge theories allows the psychologist to comprehend how therapeutic techniques effect the clients' internal constructs and also how all knowledge, both past and present, plays a role in making those connection necessary.

Bibliography

Arbib, M. (1995).
The hand book of brain theories and neural networks. Cambridge, MA: MIT press. Lusebrink, V. (1990).
Imagery and visual expression in therapy. New York: Plenum press. Martindale, C. (1991).
Cognitive psychology a neural-network approach. Belmont, CA: Brooks / Cole. McClelland, J., & Rumelhart, D. (1988).
Explorations in parallel distributed processing. Paivio, A. (1986).