Edelman, G. (1987) Neural Darwinism. Basic Books, New York, New York.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....de gnralisation de l apprentissage, des cartes topologiques peuvent se constituer, Si les rseaux de neurones superviss sont peu ralistes, d autres rseaux proposent un modle prcis et raliste du fonctionnement de nos neurones (voir par ex. les travaux de G. Edelman et son quipe, en particulier [24, 23]) Les algorithmes gntiques pour lesquels l objectif initial tait de comprendre et modliser la slection naturelle [38, 39] ont donn lieu par la suite aux systmes de classeurs [40] aux systmes cologiques [41, 30] et la simulation du systme immunitaire [4, 29] Les systmes cologiques ont montr que ....

G.M. Edelman. Neural Darwinism. Basic Books, 1987.

....when its emission has been followed by some positive consequences. This law has been observed and measured in a full range of living organisms, ranging from ies to human beings. The principle of selection by consequences is also invoked to model the formation of groups of neurons in the cortex [9], as well as the way the immune system works. Though simple to express, we think with others that the interaction of agents which behavior is selected according to this law can give rise to very complex behavioral dynamics. This law provides important adaptive capacities to organisms. Then, we ....

G.M. Edelman. Neural Darwinism. Basic Books, 1987.

....First, I start with looking at self organising neural network theories that emerged mainly from artificial intelligence. I will discuss that ordinary neural network theories will not suffice to selectively adapt languages. Secondly, the theory of neuronal group selection, introduced by Edelman [15], will be discussed as a possible candidate for the selective adaptation of language. And finally, a neuronal architecture is proposed for the implementation of the theory. 9.2 The cognitive architecture of the communicating robot In this section the architecture of the robots will be compared ....

....framework should learn (section 9.3.1) Then it is discussed why the standard neuronal network theories are not likely to be adequate of adapting languages selectionistic in the way Steels has proposed (section 9.3.2) Section 9.3. 3 briefly summarises the neuronal theory introduced by Edelman [15] as a more promising model. In section 9.3.4 a hypothetical model for language formation will be introduced. And finally, in section 9.3.5 some concluding remarks are being given. 9.3.1 What are the important features for a neural network The first sub question that I will answer is: what do we ....

[Article contains additional citation context not shown here]

Edelman, G.M. (1987) Neural Darwinism. Basic Books Inc., New York.

.... 1, or 0, but not in between. Links with a weight of 0 are e ectively pruned from the network. Thus, a structured connectionist network self organizes out of a distributed fully connected network ( gure 11) The pruning of excessive connections is common during neural development (Oppenheim, 1985; Edelman, 1987; Churchland and Sejnowski, 1992) and is theorized to improve the precision of of neural circuits (Landmesser, 1987) 6 V MAXSON In previous sections, we showed that MAXSON is a viable architecture to control an agent to act in an environment and learn from its experiences. MAXSON s modular ....

Edelman, G. M. (1987). Neural Darwinism. Basic Books, New York.

....are either close to 0, or close to 1. In the value network, the weights are close to 1, 1, or 0, but not in between. Thus, a structured connectionist network self organizes out of a distributed fully connected network. The pruning of excessive connections is common during neural development [13, 10, 7]. 6 Evolving the Learning Rules The learning algorithms in MAXSON depend on two threshold parameters ( 1 and 2 ) and their relation to the difference over time in the agent s input. The purpose of these parameters was to exclude certain perceptory inputs from the learning process. We ....

G. M. Edelman. Neural Darwinism. Basic Books, New York, 1987. 20

....against a selection criterion. The most fitted will survive and be re used in subsequent mutation recombination cycles. Genetic algorithms are often associated with classifier systems which learn classification rules as bit vectors. Genetic algorithms are a special class of selectionist systems [Edelman, 1987] which take their inspiration from biology and evolution theories. IV. The Connectionist Paradigm These methods learn by adjusting weights within a network of nodes and links [Rumelhart et al. 1987] The network structure is fixed for each learning task. This approach also relies on a large ....

Edelman, G. (1987). Neural Darwinism. Basic Books, NY.

.... error, C) maximization of information transmission ( infomax ) 24] and (D) sparseness or independence [31] Furthermore, to the design criteria listed above one should add as an important criteria ( functionality ) E) successful pattern classification, referred to and used by Edelman [11] in the context of neural Darwinism. The search for such an optimal basis leads also to the class of Projection Pursuit (PP) methods as possible candidates for universal approximation. As an example, projection pursuit 3 regression implements an additive model with univariate basis functions [14] ....

G.M. Edelman, Neural Darwinism, Basic Books, 1987.

....that one could craft such visual routines by evolving a bag of perceptual tricks whose survival is dependent on functionality and fitness. This approach, which can be directly traced to the earlier Neural Darwinism theory of neuronal group selection as a basis for higher brain function (Edelman,1987), suggests natural selection as the major force behind the automatic design of visual routines and their integration. Another possibility for evolving visual routines would employ genetic algorithms (GAs) 4 Evolutionary Computation and Genetic Algorithms The process of natural selection leads ....

....and building the neural code to capture them (Rao and Ballard, 1995) the range of imagery which could be of interest goes much beyond natural scenes. One should add within the context of face recognition an important functionality related to successful pattern classification, referred to by Edelman (1987) as neural Darwinism. Successful pattern classification amounts to the wrapper approach using subsets of features suggested by the filter approach. 9.1 Optimal Projection Axes (OPA) Sirovich and Kirby (1987) were first to apply PCA for representing face images. They showed that any particular ....

Edelman, G.M (1987). Neural Darwinism, Basic Books.

....that one could craft such visual routines by evolving a bag of perceptual tricks whose survival is dependent on functionality and fitness. This approach, which can be directly traced to the earlier Neural Darwinism theory of neuronal group selection as a basis for higher brain function (Edelman,1987), suggests natural selection as the major force behind the automatic design of visual routines and their integration. Another possibility for evolving visual routines would employ genetic algorithms (GAs) 4 Evolutionary Computation and Genetic Algorithms The process of natural selection leads ....

....and building the neural code to capture them (Rao and Ballard, 1995) the range of imagery which could be of interest goes much beyond natural scenes. One should add within the context of face recognition an important functionality related to successful pattern classification, referred to by Edelman (1987) as neural Darwinism. Successful pattern classification amounts to the wrapper approach using subsets of features suggested by the filter approach. 9.1 Optimal Projection Axes (OPA) Sirovich and Kirby (1987) were first to apply PCA for representing face images. They showed that any particular ....

Edelman, G. M. (1987). Neural Darwinism, Basic Books.

....the population approximately the same. If we denote the activation of each neuron group in v i by x i0 , x i1 , x in then, x ij = x ik , for i,j = 0, 1, n. 4) This type of lateral excitation has been used by Kohonen (1988) to acount for the formation of a cluster of active neurons and by Edelman (1989) to describe what he calls neuronal groups. # (a) # (b) v i v i x i1 x i2 x i3 Figure 1. An idealized model of cortical organization. The text describes the architecture in detail. D. Intracortical Associations The second source of activation is different for each neural group ....

Edelman, G. 1989. Neural Darwinism. Oxford University Press.

....with the environment. But many researchers in cognitive science doubt that this has anything to do with higher cognitive processes. It has been argued elsewhere that, in essence, so called high level cognition is emergent from sensory motor processes, i.e. it is grounded through embodiment [6, 11, 25, 29, 31]. For example, the embodied origins of categorization, i.e. the ability to make distinctions in the real world and to form abstract categories, and memory processes have been discussed by these authors. It turns out that if we look at the mechanisms underlying the so called high level processes ....

.... in the real world is not a computational problem, or at least not an exclusively computational one and requires that embodiment be taken into account is gaining increasing acceptance: It has been demonstrated that categorization is best viewed as a process of sensory motor coordination [6, 23]. The term sensory motor coordination which goes back to John Dewey 1896 [5] designates processes where there is a coupling of sensory and motor processes with respect to a particular purpose. For example, a robot which is turning about its own axis is not involved in a sensory motor coordination ....

[Article contains additional citation context not shown here]

Edelman, G.E. (1987). Neural Darwinism. The theory of neuronal group selection. New York: Basic Books.

....such as the taste of particular foods. But the location and perhaps the shape of these food items might vary strongly and cannot be predetermined genetically. Rather these distinctions have to be learned. The learning should be self organizing, as has been argued extensively in the literature (e.g.[8], 13] In this paper we have applied this idea to a robot that learns to distinguish between objects it should collect and other things. Classification, as it is studied in cognitive psychology, but also in computer vision and robotics, is normally treated as an information processing question: ....

G. M. Edelman. Neural Darwinism. Basic Books, New York, 1987.

....[39] Dawkins explains biological sacrificial acts with the concept of selfish genes [12] Outside the field of biology, many researchers have pointed out that evolution is important for self organization processes. Examples include the fields of chemical biology [66] and brain science [14, 15]. It is therefore natural that researchers have begun to investigate the possibility of borrowing or mimicking the mechanisms governing biological systems [61] in designing artificial systems. For example, Holland has developed a mathematical model for a natural and artificial adaptive system ....

....Problem RandLK ILK GALK att48 [5.62 5.00 95978] 5.78 5.00 94360] 5.62 5.00 97271] 0.19 1.84 3.04] 0.60 2.53 3.62] 0.53 1.58 10.85] 11 8] 21 17] 36 9] eli51 [5.03 5.00 87387] 5.03 5.00 84726] 5.05 5.00 94361] 0.23 1.15 2.58] 0.00 1.27 2.82] 0.00 1.05 3. 29] 27 10] 42 12 ] [53 14] eli76 [5.08 5.02 79788] 5.08 5.00 87393] 5.07 5.00 159884] 0.56 1.89 3.72] 0.00 1.16 2.79] 0.00 0.87 3.72] 11 9] 23 12] 24 12] kroA100 [15.18 15.00 239202] 15.15 15.00 250236] 15.18 15.00 238657] 0.33 1.34 3.98] 0.11 0.79 1.86] 0.11 0.78 1.86] 9 7] 48 10] 45 9] lin318 [301.8 300.0 ....

[Article contains additional citation context not shown here]

G. Edelman. Neural darwinism. In The Theory of Neuronal Group Selection. Basic Books, 1987.

....to model these processes is that current network systems just don t seem to capture the essence of biological systems. No matter how complex the architecture or how detailed the cellular model, there is something organic missing. Part of what is missing is provided by closing the sensorymotor loop [10] to allow real time interaction with the environment. Another critical element is a neural based control structure, e.g. attention, which dynamically reallocates computational strategies and which gates learning based changes in system properties. The core of the problem, however, is that such ....

Edelman G. Neural Darwinism. Basic Books, New York, 1987.

....the summer with Frank (Rosenblatt) in Ithaca. Subsequently Woody and HJB tried it in earnest, but on simpler and welldefined objective functions (Bremermann 1962; Bledsoe 1961b; Bledsoe and Browning 1959) The method works in principle (and is currently a popular method in Artificial Intelligence [Edelman 1987; Holland 1975] but for neural nets it turned out to be an optimization problem of extraordinary computational requirements, due to the very large numbers of synaptic weights that are involved (p. 121) Woody (Bledsoe 1962b, 1962c, 1961a) analyzed how mating, mutations, and gene interaction would ....

Edelman, G. M. 1987. Neural Darwinism. New York: Basic.

....1 We are grateful to Bill Clancey for drawing our attention to the work of John Dewey who has pointed out many of the relevant problems more than a century ago. 4 body, head, and eye muscles determining the quality of what is experienced. Dewey, 1898, pp. 137 138) In neurobiology, Edelman ([9]) has been discussing categorization from a similar perspective, by outlining the sensory motor areas that are involved in categorization. Douglas et al. 8] suggest that vision should not be viewed as passive information processing but rather as an active integrated sensorimotor event ....

....processes (grasping, pushing, turning away) While in SMC I there is one map, there is one for every sensory modality in SMC II. Categorization is achieved by a learned reentrant mapping between haptic and visual feature maps (see figure 8) The term reentry has been first introduced by Edelman ([9]) It refers to the fact that there are reciprocal connections between the feature maps. Reentry is necessary to account for the coordination of responses of modalities. When the robot explores an object there is visual and haptic stimulation. Reentry is a mechanism to correlate this perceptual ....

Edelman, G.E. (1987). Neural Darwinism. The theory of neuronal group selection. New York: Basic Books.

....operating within the brain might shape intelligent solutions to problems on the time scale of thought and action. This allows our thoughts to die instead of ourselves . More recently, selectionist approaches to learning have been studied in detail by Gerald Edelman and his collaborators (see [1] and refernces therein) The use of an evolutionary method to evolve controller architectures has been reported previously in a number of variants. Robotic controllers have, for instance, been evolved using dynamic recurrent neural nets [2] 3] Several experiments have also been performed were a ....

Edelman G. (1987) Neural Darwinism, Basic Books, New York

....operating within the brain might shape intelligent solutions to problems on the time scale of thought and action. This allows our thoughts to die instead of ourselves . More recently, selectionist approaches to learning have been studied in detail by Gerald Edelman and his collaborators (see [7] and references therein) The use of an evolutionary method to develop controller architectures has been reported previously in a number of variants. Robotic controllers have, for instance, been evolved using dynamic recurrent neural nets [4] 12] Several experiments have also been performed ....

Edelman G. (1987). Neural Darwinism., Basic Books, New York.

....on the basis of their temporal contiguity. Thus, categories develop from the real time correlations that exist accross the independent stimuli. This correlative function of reentry has been suggested to be one of the core mechanisms for the de2 velopment of categories in infants ( 12] see also [3]) A final extension of our previous work is the concept of attentional sensory motor loops which are modulated by the category specific responses of the reentrantly connected feature maps. In essense, categorization consists of breaking or enhancing the attentional sensory motor loops depending ....

....of signals of the haptic and the visual feature map by these reentrant connections forms the basic mechanism of categorization. This is in accordance with the model of categorization in infants proposed by [12] In sum, the reentrantly connected feature maps form a classification couple ([3]) A fundamental property of our model is that the feature maps are connected via modifiable feedback connections to the attention maps (see figure 4) The main idea is to link the categorical responses of the classification couple to the attentional sensory motor loop. In essence, the result of ....

G. M. Edelman. Neural Darwinism. Basic Books, New York, 1987.

No context found.

Edelman, G. (1987) Neural Darwinism. Basic Books, New York, New York.

No context found.

G. M. Edelman. Neural Darwinism. Basic Books, 1987.

No context found.

G.M. Edelman, Neural Darwinism, Basic Books, 1987.

No context found.

Edelman, G.M. (1987). Neural Darwinism. The theory of neuronal group selection. Basic Books.

No context found.

Edelman G. (1987) Neural Darwinism, Basic Books, New York

No context found.

Edelman, G. M. (1987). Neural Darwinism. New York: Basic Books.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC