Biomedical Engineering

Piezoelectric quaru crystals capable of generating ultrasonic shear waves have long been used to detect mass changes that result from deposition of metal layers, using the Sauerbrey equation to relale the observed changes in resonant frequenry to the corresponding changes in mass. More recently such crystals have been employed in

Musculoskeletal injuries are a leading occupational hazard. Apart from the physical consequences of such injuries, the social and emotional effects are serious for the injured workers, favoring a downward spiral toward chronicity. The human and economic costs of chronicity are substantial for workers, employers, compensation agencies, and the health care system. A low-cost intervention that arrests this downward spiral would be highly desirable. We decided to explore this possibility in our local setting, Edmonton, Alberta.

We have constructed a dual trap optical tweezers set-up around an inverted microscope where both the traps can be independently controlled and manipulated in all the three dimensions. Here we report our observations on rotation of red blood cells (RBCs) in a linearly polarized optical trap. Red blood cells deform and become twisted in hypertonic phosphate buffer saline and when trapped, experience an unbalanced radiation pressure force. The torque generated from the unbalanced force causes the trapped RBC to rotate. Addition of Ca ++ ions in the solution, keeping the osmolarity same, makes the cell membranes stiffer and the cells deform less. Thus the speed of rotation of the red blood cells can be controlled, as less deformation and in turn less asymmetry in shape produces less torque under the radiation pressure resulting in slower rotation at the same laser power.

Automated collection of environmental data may be accomplished with wireless sensor networks (WSNs). In this paper, a general discussion of WSNs is given for the gathering of data for educational research. WSNs enhance the scope of a researcher to include multiple streams of data: environmental, cyberdata, video, and RFID. A brief overview of the technology prepares the exploration of a setting used for outdoor learning.

Background
“Does this patient have cancer?” is a question frequently asked when confronted by patients with involuntary weight loss. The aim of this study was to assess the value of age, erythrocyte sedimentation rate (ESR), and anemia in the diagnosis of cancer as a cause of involuntary weight loss.

Methods
A retrospective study of 7850 patients admitted to the Department of Internal Medicine from January to September 2003 was performed. Especially selected were 431 patients with weight loss. Age, ESR, hemoglobin, and the discharge diagnosis were recorded.

Results
Twenty-four percent of the patients with involuntary weight loss had cancer. Age, ESR, and anemia were found not to be of value in the diagnosis of cancer (areas under the curve were 0.684, 0.690, and 0.766, respectively). When diagnostic tests for age, a high ESR, and anemia were used serially, the positive predictive value for a malignancy was 64% (CI: 27–90%); when the tests were utilized in parallel, the negative predictive value was 91% (CI: 85–100%).

Conclusions
Any patient admitted to our Department of Internal Medicine for involuntary weight loss had a 24% probability of having a malignancy. Neither age, nor ESR, nor anemia, used separately as a multilevel, diagnostic test or combined serially or in parallel, could exclude or rule in the diagnosis of cancer. However, they could increase (from 24% to 64%) or decrease (from 24% to 9%) the probability of cancer.

Natural plaque biofilms were generated in vivo on natural enamel surfaces using the device described by Robinson et al 1997. After 7 days these were recovered from the mouth and immersed in solutions of the therapeutic to be tested. After designated time intervals, devices were removed and immediately snap frozen using liquid nitrogen. After lyophilisation the accumulated biofilm was embedded in methacrylate and sectioned at 5-micron intervals from saliva interface to the enamel surface. Groups of 10 sections were pooled for determination of therapeutic content and between the 5th and 6th section a pair of 2μm sections was collected for image analysis from which biomass volume and distribution were determined. Penetration for all compounds examined was limited in 30 seconds to the outer third or so of the plaque layer. At 2 minutes concentrations in the outer regions increased but little more penetrated to the enamel surface. For fluoride, only after 30 minutes was there a semblance of equilibration with the supernatant solution. The distribution profiles obtained were compared with biomass distribution and were closely related to the surface area to mass curves suggesting that uptake may be dictated by available surface area. This may also limit penetration. From preliminary investigations, where radioactive materials were used in vitro, amine fluoride was often seen to reside at the surfaces of biomass. The much more hydrophobic riclosan, however, seemed to be located within the biomass, possibly in the region of bacterial cells.

High-resolution three-dimensional (3-D) positron emission tomography (PET) scanners with high count rate performance, such as the quad-high density avalanche chamber (HIDAC), place new demands on image reconstruction algorithms due to the large quantities of high-precision list-mode data which are produced. Therefore, a reconstruction algorithm is required which can, in a practical time frame, reconstruct into very large image arrays (submillimeter voxels, which range over a large field of view) whilst preferably retaining the precision of the data. This work presents an algorithm which meets these demands: one-pass list-mode expectation maximization (OPL-EM) algorithm. The algorithm operates directly on list-mode data, passes through the data once only, accounts for finite resolution effects in the system model, and can also include regularization. The algorithm performs multiple image updates during its single pass through the list-mode data, corresponding to the number of subsets that the data have been split into. The algorithm has been assessed using list-mode data from a quad-HIDAC and is compared to the analytic reconstruction method 3-D reprojection (RP) with 3-D filtered backprojection.

We use the \conservation of entropy" 1] to simplify the formula for the redundancy of a large class of variable-to-variable length codes on discrete, memoryless sources. This result leads to new asymptotic upper bounds on the redundancy of the \Tunstall-Hu man" code and the \Tunstall-Shannon-Fano" code.

1] Pore pressure changes are rigorously included in Coulomb stress calculations for fault interaction studies. These are considered changes under undrained conditions for analyzing very short term postseismic response. The assumption that pore pressure is proportional to faultnormal stress leads to the widely used concept of an effective friction coefficient. We provide an exact expression for undrained fault zone pore pressure changes to evaluate the validity of that concept. A narrow fault zone is considered whose poroelastic parameters are different from those in the surrounding medium, which is assumed to be elastically isotropic. We use conditions for mechanical equilibrium of stress and geometric compatibility of strain to express the effective normal stress change within the fault as a weighted linear combination of mean stress and faultnormal stress changes in the surroundings. Pore pressure changes are determined by fault-normal stress changes when the shear modulus within the fault zone is significantly smaller than in the surroundings but by mean stress changes when the elastic mismatch is small. We also consider an anisotropic fault zone, introducing a Skempton tensor for pore pressure changes. If the anisotropy is extreme, such that fluid pressurization under constant stress would cause expansion only in the fault-normal direction, then the effective friction coefficient concept applies exactly. We finally consider moderately longer timescales than those for undrained response. A sufficiently permeable fault may come to local pressure equilibrium with its surroundings even while that surrounding region may still be undrained, leading to pore pressure change determined by mean stress changes in those surroundings.

Navigator gating techniques can effectively reduce motion effects in MRI by accepting data only when the object is in a small range of positions at the cost of significantly prolonging scan time. A simultaneous multiple volume (SMV) algorithm is reported here that can substantially increase the scan efficiency while maintaining the effectiveness of motion suppression. This is achieved by acquiring different image volumes at different motion states. Initial experiments demonstrate that SMV can significantly increase the scan efficiency of navigator MRI.

Probabilistic networks (also known as Bayesian belief networks) allow a compact description of complex stochastic relationships among several random variables. They are used widely for uncertain reasoning in arti cial intelligence. In this paper, we investigate the problem of learning probabilistic networks with known structure and hidden variables. This is an important problem, because structure is much easier to elicit from experts than numbers, and the world is rarely fully observable. We present a gradient-based algorithm and show that the gradient can be computed locally, using information that is available as a byproduct of standard inference algorithms for probabilistic networks. Our experimental results demonstrate that using prior knowledge about the structure, even with hidden variables, can signi cantly improve the learning rate of probabilistic networks. We extend the method to networks in which the conditional probability tables are described using a small number of parameters. Examples include noisy-OR nodes and dynamic probabilistic networks. We show how this additional structure can be exploited by our algorithm to speed up the learning even further. We also outline an extension to hybrid networks, in which some of the nodes take on values in a continuous domain.

In this paper, we consider the general multiple target tracking problem in which an unknown number of targets appears and disappears at random times and the goal is to find the tracks of targets from noisy observations. We propose an efficient real-time algorithm that solves the data association problem and is capable of initiating and terminating a varying number of tracks. We take the data-oriented, combinatorial optimization approach to the data association problem but avoid the enumeration of tracks by applying a sampling method called Markov chain Monte Carlo (MCMC). The MCMC data association algorithm can be considered as a deferred logic since its decision about forming a track is based on the current and past observations. But, at the same time, it can be considered as an approximation to the optimal Bayesian filter. The algorithm shows remarkable performance compared to the greedy algorithm and the multiple hypothesis tracker (MHT) under the extreme conditions, such as a large number of targets in a dense environment, low detection probabilities, and a large number of false alarms.

This paper presents Markov chain Monte Carlo data association (MCMCDA) for solving data association problems arising in multiple-target tracking in a cluttered environment. When the number of targets is fixed, the single-scan version of MCMCDA approximates joint probabilistic data association (JPDA). Although the exact computation of association probabilities in JPDA is NP-hard, we prove that the single-scan MCMCDA algorithm provides a fully polynomial randomized approximation scheme for JPDA. For general multiple-target tracking problems, in which unknown numbers of targets appear and disappear at random times, we present a multi-scan MCMCDA algorithm that approximates the optimal Bayesian filter. It exhibits remarkable performance compared to multiple hypothesis tracking (MHT) under extreme conditions, such as a large number of targets in a dense environment, low detection probabilities, and high false alarm rates.

High-level actions (HLAs) lie at the heart of hierarchical planning. Typically, an HLA admits multiple refinements into primitive action sequences. Correct descriptions of the effects of HLAs may be essential to their effective use, yet the literature is mostly silent. We propose an angelic semantics for HLAs, the key concept of which is the set of states reachable by some refinement of a high-level plan, representing uncertainty that will ultimately be resolved in the planning agent's own best interest. We describe upper and lower approximations to these reachable sets, and show that the resulting definition of a high-level solution automatically satisfies the upward and downward refinement properties. We define a STRIPS-like notation for such descriptions. A sound and complete hierarchical planning algorithm is given and its computational benefits are demonstrated. This is an extended version of a paper by the same name appearing in ICAPS '07.

We study the criteria under which an object can be gripped by multi-ngered dexterous hand, assuming no static friction between the object and the ngers; such grips are called positive grips. We study three cases in detail: (i) the body is at equilibrium, (ii) the body is under some constant external force/torque, and (iii) the body is under a varying external force/torque. In each case we obtain tight bounds on the number of ngers needed to obtain grip. We also present e cient algorithms to synthesize such positive grips for bounded polyhedral/polygonal objects; the number of ngers employed in the grips synthesized by our algorithms match the above bounds. The algorithms run in time linear in the number of faces/sides.

By Fourier's theorem 1 , signals can be decomposed into a sum of sinusoids of different frequencies. This is especially relevant for hearing, because the inner ear performs a form of mechanical Fourier transform by mapping frequencies along the length of the cochlear partition. An alternative signal decomposition, originated by Hilbert 2 , is to factor a signal into the product of a slowly varying envelope and a rapidly varying fine time structure. Neurons in the auditory brainstem 3 -6 sensitive to these features have been found in mammalian physiological studies. To investigate the relative perceptual importance of envelope and fine structure, we synthesized stimuli that we call 'auditory chimaeras' , which have the envelope of one sound and the fine structure of another. Here we show that the envelope is most important for speech reception, and the fine structure is most important for pitch perception and sound localization. When the two features are in conflict, the sound of speech is heard at a location determined by the fine structure, but the words are identified according to the envelope. This finding reveals a possible acoustic basis for the hypothesized 'what' and 'where' pathways in the auditory cortex 7 -10 .

In this paper, we describe the soccer playing agent developed by our team UT Austin Villa to participate in the RoboCup 3D simulation soccer competition held in Atlanta, U.S.A., in July 2007. The 2007 competition saw the introduction of a humanoid robot in the 3D simulation league. We present an overview of this new version of the simulation league, followed by a description of our approaches in programming the humanoid robot. Our main emphasis was on developing skills such as walking and turning for the humanoid robot.

Graph shows execution time of Fortran with and without the use of shadow regions •Eliminating local copying improved performance up to factor of 2.7 Fortran versions on one processor are 3. 8 and 8.4 times faster than MATLAB Telescoping Languages Strategy: Two Phase Compilation Library Precompilation Phase •Input: MATLAB D libraries (MATLAB plus distribution types) •Perform type/distribution analysis and translate into Fortran •Create optimized variants of routines specialized for input type and context •Invoke High Performance Fortran compiler to generate communication in library routines •Automatically specializes new domain-specific libraries for different contexts and distributions •Easy to add new libraries Shifts are very expensive due to local copying •Most data in T1 and T2 was copied locally Instead, use shadow regions store a copy of off-processor data Example requires a one-element shadow region for B and for C By using shadow regions in example, •No local copying performed •No temporaries needed to hold shifted array Benefits •Fast compilation of user scripts •Performance from domain knowledge •Automatically creates new libraryaware optimizer for new libraries Base MATLAB Library Routines need to: •Be flexible for automatic specialization to different distributions and contexts •Cover functionality of MATLAB Communication generation •Minimal requirement on writer for specifying parallelism •Owner-computes rule •Relying on HPF technology to generate communication Type/distribution analysis •Best Distribution •Minimize redistribution •Obtain fastest computation •Data Movement Necessary Related Strategies •Specify parallelism by offering either: •MPI/PVM calls (too low-level) •Fixed distributions (miss optimization opportunities) •Compile the large libraries with user script (slow compile times) User Script Compilation •Input: User MATLAB D script •Perform type/distribution analysis •Determine where communication is needed •User not responsible for communication •Translate to Fortran and insert calls to data movement routines •Do not need to invoke HPF compiler •Replace general routine calls with calls to specialized variants •Obtains performance comparable to optimizing library with user script Parallel MATLAB MATLAB is a popular language for scientific applications because it is high-level and domain-specific. However, MATLAB is not a highperformance language. Scientists typically develop algorithms in MATLAB, and then need to handtranslate to a lower-level language to obtain the necessary performance. This problem is compounded by the fact that many scientific applications involve computation on large data sets that can benefit from parallel architectures. Additionally, there is limited number of expert programmers capable of optimizing to the increasingly complicated architectures. To allow scientists to easily write high-performance, parallel applications for their domains, we are developing a compiler to automatically generate parallel Fortran code from MATLAB scripts. Telescoping Languages Strategy precompiles libraries, generating variants specialized for different calling contexts, creating a library-aware optimizer. During user script compilation, library calls are quickly replaced by the best variant.

Named-entity recognition systems extract entities such as people, organizations, and locations from unstructured text. Rather than extract these mentions in isolation, this paper presents a record extraction system that assembles mentions into records (i.e. database tuples). We construct a probabilistic model of the compatibility between field values, then employ graph partitioning algorithms to cluster fields into cohesive records. We also investigate compatibility functions over sets of fields, rather than simply pairs of fields, to examine how higher representational power can impact performance. We apply our techniques to the task of extracting contact records from faculty and student homepages, demonstrating a 53% error reduction over baseline approaches.

Named-entity recognition systems extract entities such as people, organizations, and locations from unstructured text. Rather than extract these mentions in isolation, this paper presents a record extraction system that assembles mentions into records (i.e. database tuples). We construct a probabilistic model of the compatibility between field values, then employ graph partitioning algorithms to cluster fields into cohesive records. We also investigate compatibility functions over sets of fields, rather than simply pairs of fields, to examine how higher representational power can impact performance. We apply our techniques to the task of extracting contact records from faculty and student homepages, demonstrating a 53% error reduction over baseline approaches.

Much research on color representation and categorization has assumed that relations among color terms can be proxies for relations among color percepts. We test this assumption by comparing the mapping of color words with color appearances among different observer groups performing cognitive tasks: (1) an invariance of naming task; and (2) triad similarity judgments of color term and color appearance stimuli within and across color categories. Observer subgroups were defined by perceptual phenotype and photopigment opsin genotype analyses. Results suggest that individuals rely on at least two different representational models of color experience: one lexical, conforming to the culture's normative linguistic representation, and another individual perceptual representation organizing each observer's color sensation experiences. Additional observer subgroup analyses suggest that perceptual phenotype variation within a language group may play a greater role in the shared color naming system than previously thought. A reexamination of color naming data in view of these findings may reveal influences on color naming important to current theories.

There is a growing consensus that clinical evaluation of the real-world consequences of eye disease requires new performance-based tests. This is because Sueilen acuity and other common clinical tests are often poor predictors of everyday function. Ahn and Legge !(1995) Vision Research, 35, [1931][1932][1933][1934][1935][1936][1937][1938]] validated a computerized test of reading speed by showing that it provides an accurate prediction of low-vision reading performance with magnifiers. Here, we describe development of a printed-card version of the test suitable for clinical use. This printed-card test retains key design features of the validated computerized test, including the same set of sentences and display format. Data from 23 low-vision subjects showed that a very simple testing procedure using printed cards and a stop watch could be used effectively to estimate reading speed. Reading speed based on a single card was quite accurate (SD equal to about 18% of the mean) and showed no practice effects from one card to the next. Reading speeds obtained with printed cards correlated highly (r ----0.887) with those from computerized testing. We conclude that a simple test, using printed cards, can be used to obtain useful estimates of low-vision reading speed.

This is the eighth in a series of papers dealing with the role of vision in reading. In previous papers, we have evaluated the effects of stimulus and subject variables on reading rate using a drifting-text procedure. In this paper, we describe a new test of reading rate that uses static text, called the Minnesota Low-Vision Reading Test (MNread). It is microcomputer-based, and more easily set up and administered than the drifting-text procedure. It is of potential value as a standardized psychophysical test of reading and should be useful in research, clinical, and educational applications. Some types of low-vision aids rely on drifting text and others on static text. Is reading performance different for these two modes of text presentation? We measured reading rate as a function of angular character size for normal and low-vision subjects with drifting and static text. Although reading rates were highly correlated for the two modes of text presentation, normal subjects usually read static text more rapidly. The reverse was true for lowvision subjects; their reading rates for drifting text were slightly higher (average 15%) than for static text.

Fractal functions are explored as a representation for rough data in computer graphics. Two new techniques for using fractal interpolation functions to approximate rough functions and curves are introduced. The first is based on a Hough transform of fractal function transformation parameters. The second is based on previous techniques in fractal image compression. These techniques are then demonstrated on the task of recovering the parameters of a fractal function, approximating a rough function and approximating the boundary of a leaf.

Numerous techniques have been developed to perform advanced shading operations in real time. The real-time versions vary greatly from their original implementations due to the constraints of existing hardware and programming libraries. This paper introduces a grammar for articulating these modern real-time shading pipelines. We survey the existing techniques, using the grammar to classifies them into a taxonomy illustrating the commutations that differentiate the pipelines from the classical version and each other. The taxonomy follows a natural development that concludes with texture shading, which is applied to four advanced shaders to explore its versatility.

An ordered series of control poses may be interpolated by a polyscrew rigid body motion composed of a series of screws, each interpolating a pair of consecutive control poses. The trajectory of each point during screw motion is (C ∞ ). Although a polyscrew is continuous, velocities are typically discontinuous at control poses when the motion switches between screws. We obtain a smooth motion by subdividing the polyscrew. Three subdivision schemes are proposed: the interpolating 4-point subdivision, the smooth cubic B-spline subdivision, and the Jarek subdivision, an average of these two. Their implementation is trivial and their computation sufficiently fast for realtime subdivision and animation, which is particularly important for interactive motion editing and hardware support for animation.

In contrast to machined mechanical parts, the 3D shapes encountered in biomedical or styling applications contain many tubular parts, protrusions, engravings, embossings, folds, and smooth bends. It is difficult to design and edit such features using the parameterized operations or even free-form deformations available in CAD or animation systems. The Bender tool proposed here complements previous solutions by allowing a designer holding a 6 DoF 3D tracker in each hand to control the position and orientation of the ends of a stretchable virtual ribbon, which is used to grab the shape in its vicinity and to deform it in realtime, as the designer continues to move, bend, and twist the ribbon. To ensure realtime performance and intuitive control of the ribbon, we model its centerline as a circular biarc and perform adaptive refinement of the triangle-mesh approximation of the surface. To produce a natural and predictable warp, we use the initial and final shapes of the ribbon to define a one-parameter family of screw-motions. The deformation of a surface point is computed by finding its locally closest projection, or projections, on the biarc and by applying the corresponding screws, weighted by a function that decays with the distance to the projection. The combination of these solutions leads to an easy-to-use and effective tool for the direct manipulation of organic or stylized shapes.

the mapping from phenomenal states to brains. To represent movement is not to have a moving representation -presumably we represent without using the very properties that are represented. But then why should we assume in this case that representations of a certain type of movement move in the very way that is represented? We are left with a pair of troubling facts: knowing the structure of one's phenomenal states does not seem to tell us enough detail about the structure of one's brain, and knowing the structure of a represented domain does not seem to entail anything about the brain either. Suppose both that Chasles' theorem governs phenomenal states, and that we were selected to represent objects in the world as obeying Chasles' theorem. What follows about brains?

Low power implementation of digital adaptive lters for channel equalization and interference cancelling is an important aspect of wireless communications transceiver design. In this paper we propose a low power design methodology based on exploitation of the tradeo between high bit resolution and low power consumption. We focus on the widely used nite precision LMS adaptive channel equalizer for which closed form approximations can be derived for the increase in mean square error when reducing both the number of data bits and lter coe cient bits. The resulting expressions are accurate as long as the data and coe cient wordlengths are such that the weight update signal does not fall into the nite resolution deadzone of the lter. We give a su cent condition on the wordlengths which ensures that this phenomenon, known as stopping or slowdown, does not occur. Under this condition we obtain expressions for the power-optimal bit-allocation factor which determines the optimal proportion of bits to allocate to the data, the remainder being allocated to the coe cients. Numerical studies are presented for an exponential memory ISI channel with Gaussian training sequence.

A soap film is actually a thin solid fluid bounded by two surfaces of opposite orientation. It is natural to model the film using one polyhedron for each side. Two problems are to get the polyhedra for both sides to be in the same place without canceling each other out and to model triple junctions without introducing extra boundary components. We use chainlet geometry to create dipole cells and mass cells which accomplish these goals and model faithfully all observable soap films and bubbles. We introduce a new norm on chains of these cells and prove lower semicontinuity of area. A geometric version of Cartan's magic formula provides the necessary boundary coherence.

This paper outlines major challenges that we are facing in interfacing a human user with objects in the nanoworld via a haptic interface. After a review of prior efforts at haptically-enabled nanomanipulation systems, we present the current state of our nanomanipulator system. We then discuss current research issues including the direct-Z mode, force modeling, data transfer rates and the stability of the haptic interface. Results of nanomanipulation of single-walled carbon nanotubes are presented. It is our hope that the insight gained by the human user of a haptic interface to SPM will lead to scanning algorithms that can automatically adjust the SPM parameters based on the properties of the nanosample and the substrate under investigation.

Digital watermarking refers to the process of embedding a digital code into an image, video, printed document or audio media. The code should not interfere with the normal use of the media but can be recovered later. With the growing interest in haptic interactions for training and museum displays, it is only a matter of time that haptic digital media become available on the internet. For example, one will soon be able to feel the shape of Michaelangelo's David or Japan's Big Buddha by downloading a file that can be "played" on a force-feedback device. The haptic file may contain surface texture data in addition to topography so that David feels like marble and Big Buddha wood. In the meanwhile, the need for haptic digital watermarking will arise for protecting haptic media contents. This paper introduces for the first time the idea of haptic watermarking and shows one way to embed a watermark into a host surface texture signal. We also demonstrate the imperceptibility of the texture watermark in a psychophysical experiment.

The growing interest in haptic applications suggests that haptic digital media will soon become widely available, and the need will arise to protect digital haptic data from misuse. In this article, we present our study and findings on psychophysical experiments regarding human abilities to perceive a digital watermark, or hidden signal, through a haptic interface.

One of the major sources of cues in developmental learning is that of watching another person. An observer can gain a comprehensive description of the purposes of actions by watching the other person's detailed body movements. Action recognition has traditionally studied processing fixed camera observations while ignoring nonvisual information. This paper explores the dynamic properties of eye movements in natural tasks: eye and head movements are quite tightly coupled with actions. We present a method that utilizes eye gaze and head position information to detect the performer's focus of attention. Attention, as represented by eye fixation, is used for spotting the target object related to the action. Attention switches are calculated and used to segment the action sequence into action units which are recognized by Hidden Markov Models. An experimental system is built for recognizing actions in the natural task of "stapling a letter", which demonstrates the effectiveness of the approach.

Traditional supervised learning techniques do not address online learning problems such as concept drift, due to the fact that learning is offline when using these methods. Associative neural networks using Hebbian learning rules show robust performance in classification tasks involving concept drift. Biologically plausible neural networks represent a set of computational models designed to be more strongly related to biological neuron models. In this paper, we apply a biologically inspired plasticity model of synapse dynamics to a concept drift classification problem. The motivation for this method is to provide more biologically plausible networks in cognitive tasks.

A common statistical problem is that of finding the median element in a set of data. This paper presents a fast and portable parallel algorithm for finding the median given a set of elements distributed across a parallel machine. In fact, our algorithm solves the general selection problem that requires the determination of the element of rank i, for an arbitrarily given integer i. Practical algorithms needed by our selection algorithmfor the dynamic redistribution of data are also discussed. Our general framework is a distributed memory programming model enhanced by a set of communication primitives. We use efj5cient techniques for distributing, coalescing, and load balancing data as well as efficient combinations of task and data parallelism. The algorithms have been coded in SPLIT-C and run on a variety of platforms, including the Thinking Machines CM-5, IBM SP-I and SP-2, Cray Research T3D, Meiko Scientific CS-2, lntel Paragon, and workstation clusters. Our experimental results illustrate the scalability and efJiciency of our algorithms across different platforms and improve upon all the related experimental results known to the authors.

This paper summarizes the design and implementation of a parallel algorithm for state assignment of large Finite State Machines (FSMs). High performance CAD tools are necessary to overcome the computational complexity involved in the optimization of large sequential circuits. FSMs constitute an important class of logic circuits, and state assignment is one of the key steps in combinational logic optimization. The SMPbased parallel algorithm -based on the sequential program JEDI targeting multilevel logic implementation -scales nearly linearly with the number of processors for FSMs of varying problem sizes chosen from standard benchmark suites while attaining quality of results comparable to the best sequential algorithms.

There is an increasing need in modern VLSI designs for circuits implemented in high-performance logic families such as Cascode Voltage Switch Logic (CVSL), Pass Transistor Logic (PTL), and domino CMOS. Circuits designed in these noncomplementary ratioed logic families can be highly irregular, with complex diffusion sharing and nontrivial routing. Traditional digital cell layout synthesis tools derived from the highly stylized "functional cell" style break down when confronted with such circuit topologies. These cells require a full-custom, two-dimensional layout style which currently requires skilled manual design. In this work we propose a methodology for the synthesis of such complex noncomplementary digital cell layouts. We describe a new algorithm which permits the concurrent optimization of transistor chain placement and the ordering of the transistors within these diffusion-sharing chains. The primary mechanism for supporting this concurrent optimization is the placement of transistor subchains, diffusion-break-free components of the full transistor chains. When a chain is reordered, transistors may move from one subchain (and therefore one placement component) to another. We will demonstrate how this permits the chain ordering to be optimized for both intra-chain and inter-chain routing. We combine our placement algorithms with third-party routing and compaction tools, and present the results of a series of experiments which compare our technique with a commercial cell synthesis tool. These experiments make use of a new set of benchmark circuits which provide a rich sample of representative examples in several noncomplementary digital logic families.

Since their debut in 1987, snakes (active contour models) have become a standard image analysis technique with several variants now in common use. We present a framework called ''United Snakes'', which has two key features. First, it unifies the most popular snake variants, including finite difference, B-spline, and Hermite polynomial snakes in a consistent finite element formulation, thus expanding the range of object modeling capabilities within a uniform snake construction process. Second, it embodies the idea that the heretofore presumed competing technique known as ''live wire'' or ''intelligent scissors'' is in fact complementary to snakes and that the two techniques can advantageously be combined by introducing an effective hard constraint mechanism. The United Snakes framework amplifies the efficiency and reproducibility of the component techniques, and it offers more flexible interactive control while further minimizing user interactions. We apply United Snakes to several different medical image analysis tasks, including the segmentation of neuronal dendrites in EM images, dynamic chest image analysis, the quantification of growth plates in MR images and the isolation of the breast region in mammograms, demonstrating the generality, accuracy and robustness of the tool.

We present a learning technique that automatically synthesizes realistic locomotion for the animation of physics-based models of animals. The method is especially suitable for animals with highly flexible, many-degree-of-freedom bodies and a considerable number of internal muscle actuators, such as snakes and fish. The multilevel learning process first performs repeated locomotion trials in search of actuator control functions that produce efficient locomotion, presuming virtually nothing about the form of these functions. Applying a short-time Fourier analysis, the learning process then abstracts control functions that produce effective locomotion into a compact representation which makes explicit the natural quasi-periodicities and coordination of the muscle actions. The artificial animals can finally put into practice the compact, efficient controllers that they have learned. Their locomotion learning abilities enable them to accomplish higher-level tasks specified by the animator while guided by sensory perception of their virtual world; e.g., locomotion to a visible target. We demonstrate physics-based animation of learned locomotion in dynamic models of land snakes, fishes, and even marine mammals that have trained themselves to perform "SeaWorld" stunts.

Branch instructions are recognized as a major impediment to exploiting instruction level parallelism. Even with sophisticated branch prediction techniques, many frequently executed branches remain di cult to predict. An architecture supporting predicated execution may allow the compiler to remove many of these hard-to-predict branches, reducing the number of branch mispredictions and thereby improving performance. We present an in-depth analysis of the characteristics of those branches which are frequently mispredicted and examine the e ectiveness of an advanced compiler to eliminate these branches. Over the benchmarks studied, an average of 27 of the dynamic branches and 56 of the dynamic branch mispredictions are eliminated with predicated execution support.

In contrast to machined mechanical parts, the 3D shapes encountered in biomedical or styling applications contain many tubular parts, protrusions, engravings, embossings, folds, and smooth bends. It is difficult to design and edit such features using the parameterized operations or even free-form deformations available in CAD or animation systems. The Bender tool proposed here complements previous solutions by allowing a designer holding a 6 DoF 3D tracker in each hand to control the position and orientation of the ends of a stretchable virtual ribbon, which is used to grab the shape in its vicinity and to deform it in realtime, as the designer continues to move, bend, and twist the ribbon. To ensure realtime performance and intuitive control of the ribbon, we model its centerline as a circular biarc and perform adaptive refinement of the triangle-mesh approximation of the surface. To produce a natural and predictable warp, we use the initial and final shapes of the ribbon to define a one-parameter family of screw-motions. The deformation of a surface point is computed by finding its locally closest projection, or projections, on the biarc and by applying the corresponding screws, weighted by a function that decays with the distance to the projection. The combination of these solutions leads to an easy-to-use and effective tool for the direct manipulation of organic or stylized shapes.