Low-level Representation Research Articles

Using single-layered neural networks for the extraction of conjunctive rules and hierarchical classifications

Machine Learning is an area concerned with the automation of the process of knowledge acquisition. Neural networks generally represent their knowledge at the lower level, while knowledge based systems use higher level knowledge representations. The method we propose here, provides a technique which automatically allows us to extract production rules from the lower level representation used by a single-layered neural networks trained by Hebb's rule. Even though a single-layered neural network can not model complex, nonlinear domains, their strength in dealing with noise has enabled us to produce correct rules in a noisy domain.

A Further Examination of Attentional Effects in the Phonemic Restoration Illusion

Models of how listeners understand speech must specify the types of representations that are computed, the nature of the flow of information, and the control structures that modify performance. Three experiments are reported that focus on the control processes in speech perception. Subjects in the experiments tried to discriminate stimuli in which a phoneme had been replaced with white noise from stimuli in which white noise was merely superimposed on a phoneme. In the first two experiments, subjects practiced the discrimination for thousands of trials but did not improve, suggesting that they have poor access to low-level representations of the speech signal. In the third experiment, each (auditory) stimulus was preceded by a visual cue that could potentially be used to focus attention in order to enhance performance. Only subjects who received information about both the identity of the impending word and the identity of the critical phoneme showed enhanced discrimination. Other cues, including syllabic plus phonemic information, were ineffective. The results indicate that attentional control of processing is difficult but possible, and that lexical representations play a central role in the allocation of attention.

PERCEPTUAL AND CONCEPTUAL MECHANISMS IN NEGLECT DYSLEXIA:

The contribution of peripheral, data-driven effects is contrasted with conceptual, 'top-down' effects to the reading performance of 2 subjects with neglect dyslexia following a single right hemisphere lesion. Several tasks were administered, manipulating the physical, lexical or morphemic properties of the stimuli in an attempt to establish whether the attentional deficit disrupts reading at an early or late stage of processing. Both subjects were impaired at detecting elementary stimulus features on the left side of the display but were even more impaired at identifying conjoined features. One subject's performance was influenced by structural manipulations which altered the low-level representation of the stimulus. The other was less affected by structural changes of the stimuli but was influenced by the lexical and morphemic status of the words. This apparent double dissociation is interpreted as arising from a graded attentional deficit at a single locus, early in the reading process where low-level information is detected. When the deficit is not severe sufficient information may be picked up and may interact with higher order lexical knowledge to offset partially the peripheral malfunction. For a severe attentional deficit, top-down knowledge is not engaged as insufficient information is processed on the left-hand side. This hybrid view of attention provides insight into the mechanisms underlying neglect dyslexia and bears on the role of attention in normal visual processing.

Formal Specification of the Software for a Reactor Safety System

For achievement of highly dependable software the use of formal specification techniques is a must, and an associated tool support is necessary.As an example for a technique and tool environment, the specification system CIP is introduced. CIP (Computer-aided, Intuition-guided Programming) was originally developed at the Technical University of München. It supports the formal specification of software and assists the program development by stepwise transformation. A high level specification is transformed by the application of rules into lower level representation, the lowest level being the programming language level. The transformations preserve the correctness of the specifications.The use of this environment is demonstrated by an example from a safety related reactor control system. Advantages of this approach as well as drawbacks and limitations are explained.

Liberation from rectangles: A tiling method for dynamic modification of objects on raster displays

When graphics objects (also called segments) are used on raster display hardware, problems arise because of the mismatch between the high level requirements of object manipulation, and the low level representation from which the image is refreshed. This paper describes a tiling algorithm which provides a solution to the problems of damage repair and hit detection. In particular, methods for handling object priority are discussed and experimental results showing the performance of the algorithm for a number of cases are presented.

Closing the semantic gap

The behaviour of a VLSI design is viewed at different levels of abstraction during the various stages of the design process. Many authors have commented that the development of a design may be viewed as a sequence of transformations between design representations at different levels of abstraction. For instance, a design may initially be expressed with a high-level model using high-level types and functions (e.g. natural numbers and user-defined types). Subsequent design steps involve refining the high-level design model to a low-level description contains patterns for silicon implementation may be generated. The low-level description contains more information than that the original high level description (e.g. details of how numbers are represented and specific timing and area requirements). In the following discussions the lowest level of abstraction considered is a network of Boolean equations. The low-level representation is significantly different from the high level description: the high-level types and functions are defined in terms of Boolean types and functions in the We call this gap the semantic gap and the theme of this paper is using CAD tools to help bridge this gap. We discuss techniques for transforming representation of a design at, and between, different levels of abstraction and comment on how the behaviour of the low-level form can be predicted from the original high-level representation. We suggest a mechanism for generating a low-level representation of a design based on the expansion of proven recursive equations and we offer suggestions on how the testability and logic structure of a low-level design representation can be predicted from a high-level representation.

Extracting Straight Lines

This paper presents a new approach to the extraction of straight lines in intensity images. Pixels are grouped into line-support regions of similar gradient orientation, and then the structure of the associated intensity surface is used to determine the location and properties of the edge. The resulting regions and extracted edge parameters form a low-level representation of the intensity variations in the image that can be used for a variety of purposes. The algorithm appears to be more effective than previous techniques for two key reasons: 1) the gradient orientation (rather than gradient magnitude) is used as the initial organizing criterion prior to the extraction of straight lines, and 2) the global context of the intensity variations associated with a straight line is determined prior to any local decisions about participating edge elements.

MKS: A multisensor kernel system

The MKS is presented as a means for multisensor integration and data acquisition. The system has been developed in the context of a robot workstation equipped with various types of sensors utilizing three-dimensional laser range finder data and two-dimensional camera data. Specific goals that have been achieved include: developing a suitable low-level representation on raw data and/or features extracted from the raw data of the various sensors; providing a method for efficient reconfiguration of the sensor system in terms of `logical' sensors which map into physical sensors and computation; and providing a basis for high-level object modeling techniques.

Feature selection in human auditory perception

The human ability to organize and make sense out of complex arrays of information is the best example of flexible pattern recognition. Nevertheless, relatively little is known about the processes which underlie these organizational abilities. A model is proposed which describes the organizational rules or criteria employed by human listeners when comparing members of a set of complex sounds. The model assumes that feature selection is based on a Karhunen Loéve expansion of the low-level representations of sound samples. Theoretical and psychophysical analyses were performed on a set of sixteen complex sounds, revealing similar feature representations. It was concluded that the proposed model provides a reasonable first approximation to the organizational rules employed by listeners in a signal comparison task.

Spatial frequency analysis in early visual processing.

The existence of multiple channels, or multiple receptive field sizes, in the visual system does not commit us to any particular theory of spatial encoding in vision. However, distortions of apparent spatial frequency and width in a wide variety of conditions favour the idea that each channel carries a width- or frequency-related code or 'label' rather than a 'local sign' or positional label. When distortions of spatial frequency occur without prior adaptation (e.g. at low contrast or low luminance) they are associated with lowered sensitivity, and may be due to a mismatch between the perceptual labels and the actual tuning of the channels. A low-level representation of retinal space could be constructed from the spatial information encoded by the channels, rather than being projected intact from the retina.

Tutorial on programming language research

Rowe : I am going to talk a little about data abstraction from a programming language viewpoint. I suspect that what I will say in some places will be every bit as controversial amongst programming language folks, as what Dennis said was amongst database folks. Historically, programming languages evolve continually, from very low-level representations or descriptions of computations to higher-level descriptions. The idea is to express computations in a way that makes them easier to write, faster to debug, and make them survive change. All these marvelous buzzwords!

Tutorial on programming language research

Rowe : I am going to talk a little about data abstraction from a programming language viewpoint. I suspect that what I will say in some places will be every bit as controversial amongst programming language folks, as what Dennis said was amongst database folks. Historically, programming languages evolve continually, from very low-level representations or descriptions of computations to higher-level descriptions. The idea is to express computations in a way that makes them easier to write, faster to debug, and make them survive change. All these marvelous buzzwords!

Tutorial on programming language research

Rowe : I am going to talk a little about data abstraction from a programming language viewpoint. I suspect that what I will say in some places will be every bit as controversial amongst programming language folks, as what Dennis said was amongst database folks. Historically, programming languages evolve continually, from very low-level representations or descriptions of computations to higher-level descriptions. The idea is to express computations in a way that makes them easier to write, faster to debug, and make them survive change. All these marvelous buzzwords!