Machine Interpretation Research Articles

Aligning the top-level of SNOMED-CT with Basic Formal Ontology

AbstractEffective translational research requires automated analysis of large datasets collected by multiple researchers working at multiple locations. Reliable, machine interpretation of-—and reasoning with—-large datasets assembled at different times and places by different researchers requires standard representations of data. These representations are controlled, structured vocabularies also known as ontologies. By far, the most successful ontology is the Gene Ontology (GO), used by bioinformatics researchers to annotate genomics data. However, to address the phenotype side of translational research will require annotation of electronic medical record data and clinical research data with a clinical-phenotype ontology analogous to GO. One leading candidate for this ontology is SNOMED-CT (SNCT). However, GO and SNCT are incompatible representations. GO is based on an upper level ontology called Basic Formal Ontology (BFO). In this work, we aligned the upper level of SNCT with BFO to enhance its suitability for translational research. Most (14/19 or 74%) of the top-level concepts of SNCT can be fitted into the framework of BFO, but only after significant reorganization. An important concept that does not align is Clinical Finding, which is intended to comprehend diseases and signs and symptoms of disease. However, a finding of disease (epistemology) is not the same thing as a disease (ontology). This discrepancy between SNCT and BFO is important to consider further. Another key result is that children of the top-level concepts do not necessarily follow their parents into BFO, and thus one must align each SNCT concept independently. Future work is to align the next level of SNCT (345 concepts) with BFO.

Optimizing indirect branch prediction accuracy in virtual machine interpreters

Interpreters designed for efficiency execute a huge number of indirect branches and can spend more than half of the execution time in indirect branch mispredictions. Branch target buffers (BTBs) are the most widely available form of indirect branch prediction; however, their prediction accuracy for existing interpreters is only 2%--50%. In this article we investigate two methods for improving the prediction accuracy of BTBs for interpreters: replicating virtual machine (VM) instructions and combining sequences of VM instructions into superinstructions. We investigate static (interpreter build-time) and dynamic (interpreter runtime) variants of these techniques and compare them and several combinations of these techniques. To show their generality, we have implemented these optimizations in VMs for both Java and Forth. These techniques can eliminate nearly all of the dispatch branch mispredictions, and have other benefits, resulting in speedups by a factor of up to 4.55 over efficient threaded-code interpreters, and speedups by a factor of up to 1.34 over techniques relying on dynamic superinstructions alone.

A general soft method for learning SVM classifiers with [formula omitted]-norm penalty

Based on the geometric interpretation of support vector machines (SVMs), this paper presents a general technique that allows almost all the existing L 2 -norm penalty based geometric algorithms, including Gilbert's algorithm, Schlesinger–Kozinec’ s (SK) algorithm and Mitchell–Dem’ yanov–Malozemov’ s (MDM) algorithm, to be softened to achieve the corresponding learning L 1 -SVM classifiers. Intrinsically, the resulting soft algorithms are to find ε -optimal nearest points between two soft convex hulls. Theoretical analysis has indicated that our proposed soft algorithms are essentially generalizations of the corresponding existing hard algorithms, and consequently, they have the same properties of convergence and almost the identical cost of computation. As a specific example, the problem of solving ν -SVMs by the proposed soft MDM algorithm is investigated and the corresponding solution procedure is specified and analyzed. To validate the general soft technique, several real classification experiments are conducted with the proposed L 1 -norm based MDM algorithms and numerical results have demonstrated that their performance is competitive to that of the corresponding L 2 -norm based algorithms, such as SK and MDM algorithms.

Incorporating Human and Machine Interpretation of Unavailability and Rhythm Awareness Into the Design of Collaborative Applications

ABSTRACT Efficient coordination of collaboration requires sharing information about collaborators' current and future availability. We describe the usage of an awareness system called Awarenex that shared real-time awareness information to help coordinate activities at the current moment. We also developed a prototype called Lilsys that used sensors to gather additional awareness information that would help avoid disruptions when users are currently unavailable for interaction. Our experiences over time in designing and using prototypes that share awareness cues for current availability led us to identify temporal patterns that could help predict future reachability. Rhythm awareness is having a sense of regularly recurring temporal patterns that can help coordinate interactions among collaborators. Rhythm awareness is difficult to establish within distributed groups that are separated by distance and time zone. We describe rhythmic temporal patterns observed in activity data collected from users of the Awarenex prototype. Analyzing logs of Awarenex usage over time enabled us to construct a computational model of temporal patterns. We explored how to apply those patterns and model to predict future reachability among distributed team members. We discuss trade-offs in the design of collaborative applications that rely on human- and machine-interpretation of rhythm awareness cues. We also conducted a design study that elicited reactions to a variety of end-user visualizations of rhythmic patterns and investigated how well our computational model characterized their everyday routines.

Reduced Convex Hulls: A Geometric Approach to Support Vector Machines [Lecture Notes

This lecture note presents a geometric interpretation of the support vector machines (SVMs), and it introduces the reduced convex hulls (RCHs). The geometric interpretation is an alternative point of view with a clear physical meaning and paves the way for developing efficient SVM algorithms

Constraints Between Distant Lines in the Labelling of Line Drawings of Polyhedral Scenes

The machine interpretation of line drawings has applications both in vision and geometric modelling. This paper extends the classic technique of assigning semantic labels to lines subject to junction constraints, by introducing new constraints (often between distant lines). These include generic constraints between lines lying on a path in the drawing as well as preference constraints between the labellings of pairs of junctions lying on parallel lines. Such constraints are essential to avoid an exponential number of legal labellings of drawings of objects with non-trihedral vertices. The strength of these constraints is demonstrated by their ability to identify the unique correct labelling of many drawings of polyhedral objects with tetrahedral vertices. These new constraints also allowed us to deduce a general polyhedral junction constraint for the case when there is no limit on the number of faces which can meet at a junction.

Support vector machine interpretation

Decisions taken by support vector machines (SVM) are hard to interpret from a human perspective. We take advantage of a compact SVM solution previously developed, known as growing support vector classifier (GSVC), to provide interpretation to SVM decisions in terms of input space segmentation in Voronoi sections (determined by the prototypes extracted during the GSVC training method) plus rules built as a linear combination of input variables. We show by means of experiments on public domain datasets that the resulting interpretable machines have high fidelity, and an accuracy comparable to the SVM.

Putting computing in context

Context-aware computing exposes a unique tension in how information about human context is modeled and used. On the one hand, approaches that use loosely structured information have been shown to be useful in situations where humans are the final consumers of contextual information; these approaches have found favor in many CSCW applications. On the other hand, more rigidly structured information supports machine interpretation and exchange; these approaches have been explored in the ubiquitous computing community. The system presented here, dubbed Intermezzo, represents an exploration of a space between these two extremes. Intermezzo combines a loose data structuring with a number of unique features designed to allow applications to embed specialized semantic interpretations of data into the infrastructure, allowing them to be reused by other applications. This approach can enable the construction of applications that can take advantage of rich, layered interpretations of context without requiring that they understand all aspects of that context. This approach is explored through the creation of two higher-level services that provide context-enhanced session management and context-enhanced access control.

MIMes and MeRMAids: On the Possibility of Computer-aided Interpretation

One problem in humanities computing is the automatic interpretation of electronic texts. Can computers interpret texts or do they aide in the interpretation? In this paper we propose variants on the Turing Test to imagine if computer-assisted interpretation is possible. We argue that a way forward for developing models for successful machine interpreters is to work with regular expression recognition and manipulation, which is built into most programming languages in a common fashion.

Optimizing indirect branch prediction accuracy in virtual machine interpreters

Interpreters designed for efficiency execute a huge number of indirect branches and can spend more than half of the execution time in indirect branch mispredictions. Branch target buffers are the best widely available form of indirect branch prediction; however, their prediction accuracy for existing interpreters is only 2%--50%. In this paper we investigate two methods for improving the prediction accuracy of BTBs for interpreters: replicating virtual machine (VM) instructions and combining sequences of VM instructions into superinstructions. We investigate static (interpreter build-time) and dynamic (interpreter run-time) variants of these techniques and compare them and several combinations of these techniques. These techniques can eliminate nearly all of the dispatch branch mispredictions, and have other benefits, resulting in speedups by a factor of up to 3.17 over efficient threaded-code interpreters, and speedups by a factor of up to 1.3 over techniques relying on superinstructions alone.

Machine Interpretation of 2D Line Drawings as 3D Solid Objects

Machine Interpretation of 2D Line Drawings as 3D Solid Objects

Vmgen—a generator of efficient virtual machine interpreters

AbstractIn a virtual machine interpreter, the code for each virtual machine instruction has similarities to code for other instructions. We present an interpreter generator that takes simple virtual machine instruction descriptions as input and generates C code for processing the instructions in several ways: execution, virtual machine code generation, disassembly, tracing, and profiling. The generator is designed to support efficient interpreters: it supports threaded code, aching the top‐of‐stack item in a register, combining simple instructions into superinstructions, and other optimizations. We have used the generator to create interpreters for Forth and Java. Theresulting interpreters are faster than other interpreters for the same languages and they are typically 2–10 times slower than code produced by native‐code compilers. We also present results for the effects of the individual optimizations supported by the generator. Copyright © 2002 John Wiley & Sons, Ltd

Neural fuzzy preference integration using neural preference Moore machines.

This paper describes preference classes and preference Moore machines as a basis for integrating different hybrid neural representations. Preference classes are shown to provide a basic link between neural preferences and fuzzy representations at the preference class level. Preference Moore machines provide a link between recurrent neural networks and symbolic transducers at the preference Moore machine level. We demonstrate how the concepts of preference classes and preference Moore machines can be used to interpret neural network representations and to integrate knowledge from hybrid neural representations. One main contribution of this paper is the introduction and analysis of neural preference Moore machines and their link to a fuzzy interpretation. Furthermore, we illustrate the interpretation and combination of various neural preference Moore machines with additional real-world examples.

Multisensor image fusion using influence factor modification and the ANOVA methods

A new scheme is proposed for fusing multisensor images in which one image is regarded as the main image and the other the complementary; based on the evaluation of certain characteristics in the images. In effect, the scheme is used to fuse an image pair in which one image is superior to the other for interpretation in terms of higher resolution, better image quality, or having more recognizable features. Feature information is based on local statistical characteristics, which are extracted using the analysis of variance (ANOVA) method, in the framework of experimental designs. In effect, feature information from one image is used to influence the corresponding pixel values of the other image. The fused image leads to a better human and/or machine interpretation of the area of interest in the images.

Contrast Harmonic Color Doppler Left Ventriculography: Machine-Interpreted Left Ventricular Ejection Fraction Compared with Equilibrium-Gated Radionuclide Ventriculography

Background: Multi-gated acquisition (equilibrium-gated radionuclide ventriculography) (MUGA) is considered the gold standard for measuring left ventricular ejection fraction (LVEF) because it is accurate, machine interpreted, and reproducible. Echocardiographic LVEF measurements are subject to variability in image acquisition and interpretation and to the limitations of 2-dimensional (2D) versus 3-dimensional imaging. Goal: The shortcomings of traditional echocardiography may be addressed by combining multiplane 2D harmonic imaging, echocardiographic contrast, color Doppler ultrasonography, and digital image processing to create a new imaging modality: contrast harmonic color Doppler left ventriculography. Methods: We compared the accuracy of a new method for measuring LVEF that allows for machine interpretation and uses contrast-enhanced intermittent harmonic color Doppler ultrasonography (CHCD). Quantitative LVEF measurements by hand-traced harmonic 2D echocardiography, contrast-enhanced harmonic 2D echocardiography, CHCD, and machine-interpreted CHCD were compared with MUGA in 35 patients. Results: Contrast-enhanced intermittent harmonic color Doppler provided images with vivid endocardial definition in all patients, but hand-traced harmonic 2D echocardiography and contrast-enhanced harmonic 2D echocardiography had inadequate images in 9% of patients. The MUGA LVEF range was 0.09 to 0.70. All echocardiographic methods showed excellent correlation with the MUGA LVEF ( R 2 > 0.96), but the CHCD method had the best limits of agreement. Conclusions: Contrast-enhanced intermittent harmonic color Doppler LVEF correlates with MUGA at least as well as traditional noncontrasted echocardiography, but it provides diagnostic images in a greater proportion of patients. The CHCD images have vivid endocardial delineation and can be machine interpreted. (J Am Soc Echocardiogr 2000;13:368-78.)

Contaminated garbage collection

We describe a new method for determining when an object can be garbage collected. The method does not require marking live objects. Instead, each object X is dynamically associated with a stack frame M , such that X is collectable when M pops. Because X could have been dead earlier, our method is conservative. Our results demonstrate that the method nonetheless identifies a large percentage of collectable objects. The method has been implemented in Sun's Java Virtual Machine interpreter, and results are presented based on this implementation

APPLICATIONS OF SHAPE MODELING TO INTERNET COMMUNICATION

The Internet, specifically the World-Wide Web, is a radically new model of human communication and the recorded word in that the content is now directly accessible to machine interpretation. As significant as this is also that machine-readable representations of three- and higher-dimensional objects may now be processed as easily as words and two-dimensional images. The challenge is now to better integrate higher-dimensional objects into communication streams to improve the power of person-to-person communication. This paper will survey some current applications and resources for communication of three-dimensional shape information over the Internet, and suggest some new applications. We will take as motivation the symbolic processing of the "naturally-evolved" human language of mathematics, and processing "natural language" queries to on-line "dictionary" or "library" databases.

A Web-interfaced array-based mathematics course

The fascinating, often complex, often context-sensitive, classical mathematical notation presents a host of problems for machine interpretation. For those who forsee an educational environment in which interactive course material is delivered world-wide across the Web, the demand of universities (the accrediting agencies) for mastery of the classical notation is somewhat frustrating. Computer-based notations, on the other hand, can also sometimes be clumsy, confusing, inelegant and aesthetically displeasing. A solution to this impasse might seem to be yet some way off. Nevertheless, as students, rather than instructors, increasingly come to determine the choice of purchase in educational products, it might prove to be possible to acquire mastery of classical notation by routes other than the traditional ones.The USQ web-interfaced array-based first course in mathematics provides a rather different approach to the introduction of calculus. Because it is numerically based, the course largely avoids classical notation and lends itself well to distribution over the Internet. Linking the computations done in the course activities with classical notation is left to printed materials.We will briefly outline the numerical approach to the teaching of calculus, and describe how the program is delivered over the Web.

Human and Machine Interpretation of Expressions in Formal Systems

This paper uses a proof of Godels theorem, implemented on a computer, to explore how a person or a computer can examine such a proof, understand it, and evaluate its validity. It is argued that, in order to recognize it (1) as Godel's theorem, and (2) as a proof that there is an undecidable statement in the language of PM, a person must possess a suitable semantics. As our analysis reveals no differences between the processes required by people and machines to understand Godel's theorem and manipulate it symbolically, an effective way to characterize this semantics is to model the human cognitive system as a Turing Machine with sensory inputs. La logistique n'est plus sterile: elle engendre la contradicion! – Henri Poincare ‘Les mathematiques et la logique’

Video Preprocessing of Patient Environmental Data for Use in Virtual Reality Assisted Physical Therapy Systems

A variety of movement disorders, including cerebellar tremor and spasticity, may be quantified by using video-tracking techniques. Such analysis has been used to assess pharmacologic effects of antispasticity drugs and to measure performance improvements in spastic patients and in patients with cerebellar tremor by the use of corrective haptic (force) feedback. However, implementation of such techniques has been difficult in real time. Design considerations and methodology for a video preprocessor are presented. Initial results indicate that, with further development, the described approach may be used as a front end to generate therapeutically relevant positional information in real time with significantly decreased requirement for central processor commitment. Real-time output from such a preprocessor should be useable computationally to construct extensions of actual patient environments, including haptic force corridors, or permissible movement spaces, in training and assistive computerized physical therapy environments. The output of the analysis system should assist in meaningful machine interpretation of the patient's visual environment in relation to selected target-seeking behavior. Plans for refinement of the current device are discussed.