Gale L. Martin

InfoSleuth: agent-based semantic integration of information in open and dynamic environments

The goal of the InfoSleuth project at MCC is to exploit and synthesize new technologies into a unified system that retrieves and processes information in an ever-changing network of information sources. InfoSleuth has its roots in the Carnot project at MCC, which specialized in integrating heterogeneous information bases. However, recent emerging technologies such as internetworking and the World Wide Web have significantly expanded the types, availability, and volume of data available to an information management system. Furthermore, in these new environments, there is no formal control over the registration of new information sources, and applications tend to be developed without complete knowledge of the resources that will be available when they are run. Federated database projects such as Carnot that do static data integration do not scale up and do not cope well with this ever-changing environment. On the other hand, recent Web technologies, based on keyword search engines, are scalable but, unlike federated databases, are incapable of accessing information based on concepts. In this experience paper, we describe the architecture, design, and implementation of a working version of InfoSleuth. We show how InfoSleuth integrates new technological developments such as agent technology, domain ontologies, brokerage, and internet computing, in support of mediated interoperation of data and services in a dynamic and open environment. We demonstrate the use of information brokering and domain ontologies as key elements for scalability.

Recognizing hand-printed letters and digits using backpropagation learning

We report on results of training backpropagation nets with samples of hand-printed digits scanned off of bank checks and hand-printed letters interactively entered into a computer through a stylus digitizer. Generalization results are reported as a function of training set size and network capacity. Given a large training set, and a net with sufficient capacity to achieve high performance on the training set, nets typically achieved error rates of 4-5% at a 0% reject rate and 1-2% at a 10% reject rate. The topology and capacity of the system, as measured by the number of connections in the net, have surprisingly little effect on generalization. For those developing hand-printed character recognition systems, these results suggest that a large and representative training sample may be the single, most important factor in achieving high recognition accuracy. Benefits of reducing the number of net connections, other than improving generalization, are discussed.

Configuring a numeric keypad for a touch screen

Abstract This paper reports on a study comparing keying accuracy and speed for eight different numeric keypad configurations on a touch screen. Using touch-sensitive keypads displayed on a computer terminal, operators entered numbers presented to them through a speech synthesizer. Dependent measures collected were keying rates, errors, and the x- and y-dimension standard deviations from the centre point of the key. The primary finding was that keypads with square keys resulted in improved speed and a higher degree of accuracy than do keypads with regular keys (either a long horizontal dimension or a longer vertical dimension). Spread-out versions of the keypads (inter-key spacing = 1·3 cm) did not yield superior performance compared with compressed versions (inter-key spacing = 0·6 cm).

INTEGRATED SEGMENTATION AND RECOGNITION THROUGH EXHAUSTIVE SCANS OR LEARNED SACCADIC JUMPS

This paper advances two approaches to integrating handwritten character segmentation and recognition within one system, where the underlying function is learned by a backpropagation neural network. Integrated segmentation and recognition is necessary when characters overlap or touch, or when an individual character is broken up. The first approach exhaustively scans a field of characters, effectively creating a possible segmentation at each scan point. A neural net is trained to both identify when its input window is centered over a character, and if it is, to classify the character. This approach is similar to most recently advanced approaches to integrating segmentation and recognition, and has the common flaw of generating too many possible segmentations to be truly efficient. The second approach overcomes this weakness without reducing accuracy by training a neural network to mimic the ballistic and corrective saccades (eye movements) of human vision. A single neural net learns to jump from character to character, making corrective jumps when necessary, and to classify the centered character when properly fixated. The significant aspect of this system is that the neural net learns to both control what is in its input window as well as to recognize what is in the window. High accuracy results are reported for a standard database of handprinted digits for both approaches.

Centered-object integrated segmentation and recognition of overlapping handprinted characters

Visual object recognition is often conceived of as a final step in a visual processing system, First, physical information in the raw image is used to isolate and enhance to-be-recognized clumps and then each of the resulting preprocessed representations is fed into the recognizer. This general conception fails when there are no reliable physical cues for isolating the objects, such as when objects overlap. This paper describes an approach, called centered object integrated segmentation and recognition (COISR), for integrating object segmentation and recognition within a single neural network. The application is handprinted character recognition. The approach uses a backpropagation network that scans a field of characters and is trained to recognize whether it is centered over a single character or between characters. When it is centered over a character, the net classifies the character. The approach is tested on a dataset of handprinted digits and high accuracy rates are reported.

The InfoSleuth Project

The InfoSleuth Project at MCC [7, 9, 8, 1] is developing and deploying technologies for finding information in corp~ rate networks and in external networka, such as networks baaed on the emerging National Information Infrastructure. InfoSleuth is baaed on MCC’S previously developed Carnot technology [2, 6, 10], which was successfully used to integrate heterogeneous information resources. The Carnot project developed semantic modeling techniques that enable description of the information resources and pioneered the use of agents to provide interoperation among autonomous systems. The InfoSleuth Project investigates the use of Carnot technologies in a dynamically changing environment, such as the Internet, where there is no formal control of the registration of new information sources and the identities of the resources to be used may be unknown at the time the application is developed. InfoSleuth deploys semantic agents [9, 5, 3] that carry out coordinated searches and cooperate with each other to merge the retrieved data into understandable information. The project is developing technologies to support mediated interoperation of data and services over information networks in a dynamically changing environment, including:

InfoSleuth: agent-based system for data integration and analysis

InfoSleuth is an agent-based system that automates the gathering and analysis of dynamic, distributed data accessible over the web. To gather data, a user specifies an SQL query which references elements from a common domain ontology (i.e., terminology). Agents then locate resources that have advertised having data relevant to these elements and translate the ontology-based query into queries referencing elements in the different schemas of the identified local databases. Other agents then integrate the results returned from these multiple resources and express them in terms of the common ontology. The large amount of data returned may be overwhelming, and so analysis agents serve to filter and interpret it. InfoSleuth is implemented in Java, and includes a common agent shell and specializations of the shell. The system was developed within a research environment over the course of 5 years and is now being hardened for commercial applications.

Human performance evaluation of digitizer pucks for computer input of spatial information

Two experiments were performed to measure and find ways of improving the accuracy with which people can enter spatial coordinates into a computer with a digitizer puck. A digitizer puck is similar to a mouse; however, rather than serving to position a cursor on a display screen, the puck is used to enter the spatial coordinates of points in a hard-copy document such as a map, photograph, or image representation of text. The first experiment studied the effect of the type of optical sight provided on the puck. The second experiment examined the effect of using a 2.5-power magnifier. Varying the type of optical sight used did not affect accuracy, but use of the magnifier did significantly improve accuracy. The study also suggested some initial estimates of the accuracy levels obtainable by most people. Most of the subjects were accurate to within ± 0.125 mm to ± 0.250 mm of the true target position. Magnification drives the error value down to the lower ends of this range.