D.H. Ballard, Cortical connections and parallel processing: Structure and function, Behavioral and Brain Sciences, 9, 67-120, 1986.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....frequency, low frequency tuned cells tend to be more reliable for the purposes of motion velocity estimation. Second, model C2M may have benefited from the fact that complex cells with large, overlapping receptive fields provide a high resolution coarse code of the spatiotemporal space [10] [12]. This code could provide model C2M with accurate information as to the location of the moving object at each moment in time. For example, the activities of the population of these cells may have coded with high accuracy the fact that the moving object was at location at time and at location ....

Ballard, D.H. (1986) Cortical connections and parallel processing: Structure and function. Behavioral and Brain Sciences, 9, 67-120.

....representation, not a matrix corresponding to the 2D projection of the world into which the retinotopic information is deposited. Feldman uses value units to induce a mapping between the two frames (much as is done with the Hough transform mapping from images to parameter spaces; see Ballard, 1986). This differs from the present approach, which does not map the entire retinotopic representation onto some global space at all, but only provides indexes to selected FINSTed features, and cross bindings to a descriptive symbolic representation. 7 A Model of Spatial Indexing processing. ....

Ballard, D.H. (1986). Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9, 67-120.

....or the state of a unit. Moreover, units may be connected to outside environments. In this case the units are sometimes assumed to respond to a narrow range of combinations of parameter values and are said to have a certain receptive field in parameter space. These are called value units (Ballard, 1986). In some versions of Connectionist architecture, environmental properties are encoded by the pattern of states of entire populations of units. Such coarse coding techniques 2 are among the ways of achieving what Connectionists call distributed representation . The term Connectionist model ....

....prove to be well founded; then that would be incompatible with the assumption that the cognitive architecture of the capacities acquired is Connectionist. On Brain style modeling The relation of Connectionist models to neuroscience is open to many interpretations. On the one hand, people like Ballard (1986), and Sejnowski (1981) are explicitly attempting to build models based on properties of neurons and neural organizations, even though the neuronal units in question are idealized (some would say more than a little idealized: see, for example the commentaries following the Ballard (1986) paper) ....

[Article contains additional citation context not shown here]

Ballard, D.H. (1986). Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9, 67-120.

....(I) T In our experiments we use N I = 10. Thus, for a value of I = 0:5, the k = 5 and k = 6 components will fire strongly while the rest of the components will be quite small. This representation is an example of a value encoding, as opposed to a variable encoding, in the sense discussed in [2]. In this manner, an intensity of 0:1 is held in as high a regard as an intensity of 0:9; each has its own place on the number line, and the former should not be viewed as a weak or low magnitude version of the latter. Value encodings in this sense have been widely accepted in the context of ....

D.H. Ballard. Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9:67--120, 1986.

....The hypothesized involvement of MDS in perception does not mean that the perceptual system actually carries out an MDS computation to recover the distal contrasts explicitly. The activity pattern of units selective for the relevant prototypes constitutes a distributed coarse coded representation (Ballard, 1986) of the desired contrasts, which need not be converted into a localized form to be put to use. It should be stressed that, as an explanatory device in theory of perception, coarse coding is only serviceable insofar as it is invoked together with the notion of hyperacuity. In other words, what may ....

Ballard, D. H. (1986). Cortical connections and parallel processing: structure and function. Behavioral and Brain Sciences, 9:67--120.

....A CORTICAL COLUMN 2.1 INTRODUCTION The cortical column, the module concept , has been proposed as an anatomical entity by Szentagothai in [32] Later, Mountcastle in [22] and Zeki and Shipp in [36] enriched the hypothesis by describing the functional context of the cortical column. Ballard in [2] was the first to propose the model that clearly identifies three distinct parts of the column: the upper, the intermediate and the lower divisions. Burnod in [14] proposed the concepts of a columnar call and action tree and the use of the trees to explain the functionality of the cortex. A ....

Ballard, D. H. (1986), Cortical connections and parallel processing: structure and function, Behavioral and Brain Sciences.

....to output. We can distinguish between processors for which output is proportional to net input, such as those above, from those that are sensitive to only a particular range of values. Ballard refers to the former type of unit as an integration device, whereas the latter he terms a value unit [1]. In the present context we will refer to any unit for which unit activity is not proportional to its net input as non monotonic. Dawson and Schopflocher have developed a connectionist architecture based on the concept of a value unit [5] A network of value units is a feed forward multi layer ....

D. H. Ballard. Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9:67--120, 1986.

....had spherical receptive fields in which the variance determined the field s width. The variance of the Gaussian units was set so that there was considerable overlap of the units receptive fields. Consequently, the units provided the networks with a coarse coded representation of the retinal array (Ballard, 1986; Hinton, 1981; Hinton, McClelland, and Rumelhart, 1986) An important feature of such codes is that their resolution depends on the degree of overlap. High resolution codes are formed by units with large, overlapping receptive fields whereas units with small, less overlapping receptive fields ....

....receptive fields (oe 2 = 0:8) and provide a relatively low resolution code, the units on the right have large receptive fields (oe 2 = 1:8) and provide a high resolution code (for graphical purposes, the receptive field of a 12 unit is shown as a circle whose radius is 2oe) 1 . Analyses by Ballard (1986) and Hinton (1981) provide a quantitative understanding of the relationship between receptive field size and resolution in the case of binary units that each become active when a stimulus falls within its receptive field. If D is the diameter of a unit s receptive field, k is the dimensionality of ....

[Article contains additional citation context not shown here]

Ballard, D. H. (1986) Cortical connections and parallel processing: Structure and function.

....that this foveocentric structure plays a key role in generating the visual response properties of many cells in MSTd. While the primary projection to MSTd comes from MT (Maunsell and Van Essen, 1983) a second route by which optic flow information may reach MSTd is the tectopulvinar pathway (c.f. Ballard, 1987). This pathway bypasses the the primary visual cortex altogether, passing motion signals from the superior colliculus through the pulvinar to MT (Standage and Beneveto, 1983) 0 1 2 3 4 5 6 40 80 120 160 10 20 LOG POLAR DIRECTION TUNING s (DEG) Figure 17: Heading sensitivity is unaffected by ....

....(1997) A neural model of high level motion processing: Line motion and formotion dynamics. Vision Res 37:3037 3059. Baloch A, Grossberg S, Mingolla E, Nogueira CAM (1999) A neural model of first order and second order motion perception and magnocellular dynamics. J Optical Soc Amer A, 16:953978. Ballard D (1987) Cortical connections and parallel processing: Structure and function. In: Vision, Brain, and Cooperative Computation (Arbib M, Hanson A, eds. pp. 563 621. Cambridge: MIT Press. Bradley D, Maxwell M, Andersen RA, Banks MS, Shenoy K (1996) Mechanisms of heading perception in primate visual ....

[Article contains additional citation context not shown here]

Ballard D (1987) Cortical connections and parallel processing: Structure and function. In: Vision, Brain, and Cooperative Computation (Arbib M, Hanson A, eds.), pp. 563-621.

....Figure 3: Learning curves for the 26 18 9 network on the what and where tasks using random and blocked training. 100 75 50 25 0 0 20 40 60 80 100 Epoch Percent Correct random training blocked training standard error of the mean at epoch 50 = 4. 25 Figure 4: Learning curves for the 26 36 9 network on the what and where tasks using random and blocked training. nificantly faster with random training than with blocked training (at epoch 50, the difference between the performance with random and blocked training is statistically significant at the p 0.01 level (t = 7.89) ....

....faster with random training than with blocked training (at epoch 50, the difference between the performance with random and blocked training is statistically significant at the p 0.01 level (t = 7. 89) Similar to the 26 18 9 network described above, temporal crosstalk attenuates the 26 36 9 network s rate of learning in blocked training. Therefore, we cannot conclude that an abundance of hidden units makes a network more robust in the presence of temporal crosstalk. The modular architecture that we applied to the what and where tasks, shown in 36 Matrix 5 X 5 Task bit 9 5 ....

[Article contains additional citation context not shown here]

Ballard, D.H. (1986) Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9, 67--120.

....100 million might be a reasonable ballpark figure. If we limit ourselves to one object at a time, however, and the object s position, orientation, and scale are computed first, then the number of object instantiations that have to considered at once is only o, or a number more like 10,000. Ballard (1986) and Tsotsos (1990, 1991) have presented computational complexity analyses of this sort to argue that the combinatorics of vision require some type of attentional selection to reduce the number of possibilities that need to be considered, and that attention can be particularly beneficial when ....

Ballard, D. H. (1986). Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9, 67--120.

....a greater understanding of the approach and progress of the network in learning. By comparison, multi layer networks are very much black box mechanisms. 3. Physical constraints. In the human brain, there is a limit to the number of neurons that can be accommodated in the available space. Ballard[27] has CHAPTER 4. UNCONVENTIONAL AI MODELS 23 hypothesised that to represent a space of dimension k requires N k =D k Gamma1 neurons, where N is the number of just noticeable differences in each dimension of the space, and D is the diameter of the receptive field of each neuron. To accommodate ....

Ballard, D. H. Cortical Connections and Parallel Processing: Structure and Function, Behavioural and Brain Sciences, Vol. 9. pp 67-120.

....thesis since the movies which the central system will learn to categorize are sequences of static frames. 3. 1 Overview Connectionism is a neurally inspired style of computation in which many simple interconnected processors cooperatively perform some computational task [Feldman and Ballard, 1982; Ballard, 1987a] This field, also known as the field of neural networks, has had a profound impact in recent years on the way in which computation is conceived of, primarily by providing an alternative to the classic von Neumann model of a program residing in memory dictating the actions of a central processing ....

Dana H. Ballard, "Cortical Connections and Parallel Processing: Structure and Function," In Michael A. Arbib and Allen R. Hanson, editors, Vision, Brain, and Cooperative Computation. MIT Press, 1987.

....thesis since the movies which the central system will learn to categorize are sequences of static frames. 3. 1 Overview Connectionism is a neurally inspired style of computation in which many simple interconnected processors cooperatively perform some computational task [Feldman and Ballard, 1982; Ballard, 1987a] This field, also known as the field of neural networks, has had a profound impact in recent years on the way in which computation is conceived of, primarily by providing an alternative to the classic von Neumann model of a program residing in memory dictating the actions of a central processing ....

Dana H. Ballard, "Cortical Connections and Parallel Processing: Structure and Function," In Michael A. Arbib and Allen R. Hanson, editors, Vision, Brain, and Cooperative Computation. MIT Press, 1987.

....general problem solver. 1.0 Introduction The cortical column, the module concept , has been proposed as an anatomical entity by Szentagothai [1] Later, Mountcastle [2] and Zeki and Shipp [3] enriched the hypothesis by describing the functional context of the cortical column. Ballard in [4] was the first to propose the model that clearly identifies three distinct parts of the column: the upper, the intermediate and the lower divisions. Burnod in [5] proposed the concepts of a columnar call and action tree and the use of the trees to explain the functionality of the cortex. A ....

Ballard, D. H., "Cortical connections and parallel processing: structure and function", Behavioral and Brain Sciences, 1986.

....Psychophysical evidence supports the view that the interaction and inter communication between the visual neurons may play an important role in enabling useful visual behaviour The work is funded by the ESPRIT EP2152 (VIEWS) project. to emerge effectively and efficiently from simple neurons [20, 26, 3, 14, 13]. Braitenberg [5] has suggested a similar principle for assembling artificial behaviours whilst, more recently, Brooks [6, 7] has demonstrated the importance of building integrated architectures for robots with task oriented behaviours. Many researchers have shown the potential for building vision ....

D. Ballard. Cortical Connections and Parallel Processing: Structure and Function, The Behavioral and Brain Science. Cambridge University Press, Cambridge, England, 1985.

....on a cortical column 6.1 Introduction The cortical column, the module concept , has been proposed as an anatomical entity by Szentagothai in [55] Later, Mountcastle in [38] and Zeki and Shipp in [68] enriched the hypothesis by describing the functional context of the cortical column. Ballard in [3] was the first to propose the model that clearly identifies three distinct parts of the column: the upper, the intermediate and the lower divisions. Burnod in [14] proposed the concepts of a columnar call and action tree and the use of the trees to explain the functionality of the cortex. A ....

....of the first and second level indices are clearly defined. Neuromorphic Distributed General Problem Solvers, Andrzej Bieszczad 228 Systems and Computer Engineering, Carleton University Figure 80 illustrates the two tier organization of the topology preserving maps of the cortex (after Ballard [3]) Two regions with the first level index spaces A and B contain several regions with the second level index spaces. Some of those sub regions, namely Y and Z, are repeated in A and B. This implies that a sensation that can be indexed by Y is a component of two different compound perceptions. As ....

[Article contains additional citation context not shown here]

Ballard, D. H. (1986), Cortical connections and parallel processing: structure and function, Behavioral and Brain Sciences.

....at minute from the first sample of a day. The target output of the neural net is a prediction of the change in the series value Delta p minutes in the future, i.e. s( Delta s t Delta p ) Gamma s( Delta s t) Each of the input and output quantities is represented using a variable encoding (Ballard, 1986), which means that the input and output activities are monotonic in the represented quantity. This is the obvious encoding; I mention it simply because connectionist approaches to representation Mozer 13 (e.g. Ballard, 1986; Smolensky, 1990) offer a range of alternatives that could be ....

....input and output quantities is represented using a variable encoding (Ballard, 1986) which means that the input and output activities are monotonic in the represented quantity. This is the obvious encoding; I mention it simply because connectionist approaches to representation Mozer 13 (e.g. Ballard, 1986; Smolensky, 1990) offer a range of alternatives that could be fruitfully explored, but I have not done so in the present work. The activity of each input unit was normalized to have a mean of 0.0 and variance 1.0 over the training set. This leads to a better conditioned search space for learning ....

Ballard, D. H. (1986). Cortical connections and parallel processing: Structure and function. The Behavioral and Brain Sciences, 9, 67--120.

....learning in the presence of modularity, but it would be much nicer to actually learn the divisions of a task. I wonder if it would be possible to create a model that expressed modularity using several POMDP s to model different aspects of the world. This idea is related to Ballard s value units [ Ballard, 1986 ] a technique that used several separate neural networks to cut down the exponential explosion of units required to represent all possible combinations of multidimensional values. Drescher has presented a good example of learning modularity [ Drescher, 1991 ] The finite state machine model of ....

Dana H. Ballard. Cortical connections and parallel processing: Structure and function. Behavioral and Brain Sciences, 9(1), March 1986.

....explicit representation of coordinates in the activity values has two shortcomings. First, biological visual systems do not appear to encode coordinates explicitly. Instead, they represent positions by the locations of active cells in retinotopic maps, i.e. by value unit encoding (Barlow, 1972; Ballard, 1987). Second, explicit encoding may be problematic for hardware implementation. The operational range and precision of devices 1 determine the maximum number of locations that can be encoded, which in turn limits the size of the largest map. Judging by the number of neurons in biological maps and ....

Ballard, D. H. (1987). Cortical connections and parallel processing: Structure and function.

....logistic have an order of 1 [74] and are able to carve a pattern space into two distinct regions (see Figure 10) Thus, networks using this form of activation function can solve linearly inseparable problems without any hidden units. These networks have been termed Integration Devices by Ballard [6], and generic PDP nets by Anderson and Rosenfeld [3] Figure 10: A monotonic activation function such as the logistic divides a pattern space into two distinct regions The power of these simple units emerges when they are connected together to form a network, or multi layer perceptron (MLP) ....

....an activation function based upon neurophysiological evidence. Evidence from single unit recordings (that is, record the output of the neuron with respect to its input) suggests that there are at least two functionally different types of neurons in the brain in regards to their output encodings [6]. This can be illustrated by comparing the recordings from neurons that function as a basic part of the oculomotor system to neurons in the visual areas of the cortex. The first type of neurons for example, those in the servo system controlling eye movement have linear outputs whose firing ....

[Article contains additional citation context not shown here]

D. H. Ballard. Cortical connections and parallel processing: Structure and function. Behavioral and Brain Sciences, 9:67--120, 1986.

....current scene coordinates with respect to the retinal frame. There is an asymmetry between these frames, in that the former needs to keep track of multiple targets, whereas the latter needs only keep track of a single data point, namely, the relation of the scene to retinal origin. As discussed in [Ballard, 1986] , there are two basic kinds of encoding that can be employed by neurons. For multiple targets, a value unit encoding can be used, wherein the neuron s receptive field is localized and the peak firing rate signals a specific location. For single targets, a variable unit encoding can be used, ....

D.H. Ballard. Cortical connections and parallel processing: Structure and function. Behavioral and Brain Sciences, 9(1):67--120, 1986.

....memory the references to the items therein have to be changed with the requirements of the ongoing computation. The strategy of copying that was used as a straw man in the silicon example is even more implausible here, as most neurons in the cortex exhibit a form of place coding [Barlow, 1972, Ballard, 1986] that cannot be easily changed. Thus it seems that at the one third second time scale, ways of temporarily binding huge numbers of neurons and changing those bindings must exist. That is, the brain must have some kind of pointer mechanism. 5 1.3 Outline The focus of this paper is to explain why ....

Ballard, D. (1986). Cortical connections and parallel processing: Structure and function. Behavioral and Brain Sciences, 9(1):67--120.

No context found.

D.H. Ballard, Cortical connections and parallel processing: Structure and function, Behavioral and Brain Sciences, 9, 67-120, 1986.

No context found.

D. H. Ballard. Cortical connections and parallel processing: structure and function. Behavioral and Brain Sciences, 9:67--120, 1986.

No context found.

D. H. Ballard. Cortical connections and parallel processing: structure and function. Behavioral and Brain Sciences, 9:67--120, 1986.

No context found.

D. H. Ballard. Cortical connections and parallel processing: structure and function. Behavioral and Brain Sciences, 9:67--120, 1986.

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