Reinhold C. Mann

Restoration of piecewise-constant images by mean-field annealing

An algorithm is described that removes the noise from images without causing blurring or other distortions of edges. The problem of noise removal is posed as a restoration of an uncorrupted image, given additive noise. The restoration problem is solved by using a new minimization strategy called mean-field annealing (MFA). An a priori statistical model of the image is chosen that drives the minimization toward solutions that are locally homogeneous. The strategy for MFA is derived, and the resulting algorithm is discussed. Applications of the algorithm to both synthetic images and real images are presented.

A neural network algorithm for the multiple traveling salesmen problem

We developed an efficient neural network algorithm for solving the Multiple Traveling Salesmen Problem (MTSP). A new transformation of the N-city M-salesmen MTSP to the standard Traveling Salesmen Problem (TSP) is introduced. The transformed problem is represented by an expanded version of Hopfield-Tanks neuromorphic city-position map with (N + M-1)-cities and a single fictitious salesmen. The dynamic model associated with the problem is based on the Basic Differential Multiplier Method (BDMM) [26] which evaluates Lagrange multipliers simultaneously with the problems state variables. The algorithm was successfully tested on many problems with up to 30 cities and five salesmen. In all test cases, the algorithm always converged to valid solutions. The great advantage of this kind of algorithm is that it can provide solutions to complex decision making problems directly by solving a system of ordinary differential equations. No learning steps, logical if statements or adjusting of parameters are required during the computation. The algorithm can therefore be implemented in hardware to solve complex constraint satisfaction problems such as the MTSP at the speed of analog silicon VLSI devices or possibly future optical neural computers.

Mean-field approximation minimizes relative entropy

We derive the mean-field approximation from the information-theoretic principle of minimum relative entropy instead of by minimizing Peierls’s inequality for the Weiss free energy of statistical physics theory. We show that information theory leads to our statistical mechanics procedure. As an example, we consider a problem in binary image restoration. We find that mean-field annealing compares favorably with the stochastic approach.

An artificial intelligence approach to DNA sequence feature recognition

The ultimate goal of the Human Genome project is to extract the biologically relevant information recorded in the estimated 100,000 genes encoded by the 3 x 10(9) bases of the human genome. This necessitates development of reliable computer-based methods capable of analysing and correctly identifying genes in the vast amounts of DNA-sequence data generated. Such tools may save time and labour by simplifying, for example, screening of cDNA libraries. They may also facilitate the localization of human disease genes by identifying candidate genes in promising regions of anonymous DNA sequence.

Learning algorithms for feedforward networks based on finite samples

We present two classes of convergent algorithms for learning continuous functions and regressions that are approximated by feedforward networks. The first class of algorithms, applicable to networks with unknown weights located only in the output layer, is obtained by utilizing the potential function methods of Aizerman et al. (1970). The second class, applicable to general feedforward networks, is obtained by utilizing the classical Robbins-Monro style stochastic approximation methods (1951). Conditions relating the sample sizes to the error bounds are derived for both classes of algorithms using martingale-type inequalities. For concreteness, the discussion is presented in terms of neural networks, but the results are applicable to general feedforward networks, in particular to wavelet networks. The algorithms can be directly adapted to concept learning problems.

GRAIL: an integrated artificial intelligence system for gene recognition and interpretation

The development of an integrated artificial intelligence system, GRAIL (gene recognition and analysis Internet link) is described. This system uses a combination of a multi-sensor/neural network, expert system, and parallel search tools to recognize and interpret genes in DNA sequences. A simple electronic mail (E-mail) interface makes the system accessible through Internet. The strength of the system in recognizing and interpreting genes in DNA sequences and the simple E-mail interface have already attracted more than 150 users. The success of the system is largely due to the multi-sensor/neural network approach and the integration of several AI tools. The modular development and flexible framework have made it easier to incorporate new knowledge and tools into the existing system.<<ETX>>

The development of an intelligent nuclear maintenance robot

The US Department of Energy has provided support to four universities and the Oak Ridge National Laboratory (ORNL) to pursue research leading to the development and deployment of robotic system(s) for advanced nuclear power stations. The scope of the program and the R&D effort contributed by ORNL are reported.<<ETX>>

Use of multiparameter analysis to quantitate hematological damage from exposure to a chemical (ethylene oxide)

This study was designed to test the value of a multiparameter approach in evaluating perturbations in bone marrow and peripheral blood elements of mice exposed to ethylene oxide (EtO). Mice exposed to 255 ppm EtO for 5 h/d were removed for analysis after 1, 2, 8, and 14 d (sequential exposure) and 4, 6, 8, and 10 wk (5 d/wk). Prior to sacrifice, blood was removed from the orbital sinus for blood cell counts, hemoglobin determination, and hematocrit. A blood film was made for differential leukocyte counts. Bone marrow was flushed from femurs and tibias and counted, and aliquot were used for stem-cell assay (CFU-S) or flow cytometry (FCM) analysis. One aliquot of marrow was stained with propidium iodide for cell-cycle analysis and another was reacted with fluorescein-conjugated monoclonal antibody for B-cell analysis. The preparations were analyzed for forward and 90 degrees scatter and fluorescence on an Ortho 50H cytofluorograph. Perturbations of peripheral leukocytes occurred after one exposure. After multiple exposures, hematocrit, red-cell number, and hemoglobin were generally depressed, with transient compensatory bursts, and bone marrow cellularity and CFU-S were below normal. However, white-cell numbers fluctuated dramatically during the exposure period. There was a shift in differential toward granulocytes, at times resulting in severely depressed numbers of lymphocytes in the peripheral blood. The FCM analysis showed an early depletion of granulocytes in the bone marrow followed by replacement and a relative lymphocyte deficit, especially pronounced at 10 wk. The B-cell changes reflected general lymphocyte perturbations. Shifts in numbers of cells in S and G/M were observed, consistent with a moderate bone marrow response to cell loss.

Multi-Sensor Integration Using Concurrent Computing

This paper addresses methods for high and low level multi-sensor integration based on maintaining consistent labelings of features detected in different sensor domains. Implementation in a concurrent computing environment is discussed. Keywords: Multi-Sensor Integration, Sensor Fusion, Consistent Labeling, Markov Random Field, Concurrent Computing, Hypercube, Simulated Annealing.

AUTOMATED ANALYSIS OF DIGITAL IMAGES GENERATED BY TWO-DIMENSIONAL GEL ELECTROPHORESIS

Two-dimensional gel electrophoresis is a method to separate proteins according to their molecular weights and mobilities in an electric field at a certain pH value. The presence of a radioactively or fluorescently labeled protein in a preparation is indicated by a spot that is produced on a film with the intensity of the spot indicative of the amount of protein. This article presents algorithms that accomplish automated quantitative analysis of digitized gel images. The limits of currently available analysis systems are discussed. The methods are implemented by software on a commercially available digital image processor that uses a minicomputer host. 13 refs., 4 figs., 3 tabs.