Parallel implementation of neural networks

This paper presents systematic methods, based on graph theoretic approach, for mapping of neural networks onto mesh connected SIMD arrays. The methods are applicable to a large class of multilayer network models, which can be represented in terms of sparse matrix vector operations. The class of computers, that the mappings are suitable for, encompasses most of the experimental and commercial mesh-connected SIMD arrays of processors. There are three methods described in the paper, one for the case of a processor array, which is larger or equal to the network size and two for the partitioned case, i.e. array smaller than the input data size. The methods are illustrated on an example of a multilayer perceptron with back-propagation learning, which consists ofn nuerons ande synaptic connections. For the first method, the processor array is assumed to be of sizeN×N, whereN ² n+e, and the required local memory of processors is limited to only a few registers. The implementation of a single iteration of a recall phase according to the method requires 24(N-1) shifts. For this method we have developed a software tool, which generates a sequence of pseudo instructions, such as elemental data shift and arithmetic operations, that implement a given neural network on a given size processor array. For the two partitioned methods, the processor array is of sizeP×P, whereP ²n+e, and the local memory in the processors is of sizeO(K). The faster of the two methods requiresO(N ³/P ³ K) time for an iteration of the recall or learning phase.This research was supported in part by the National Science Foundation under grant MIP-8714689 and IRI-9145810.Preliminary versions of the results contained in this paper appear in the International Conference on Applications-Specific Array Processors 1990 and the IEEE Workshop on VLSI Signal Processing 1990.     

Hybrid genetic-paired-permutation algorithm for improved VLSI placement

This paper addresses Very large-scale integration (VLSI) placement optimization, which is important because of the rapid development of VLSI design technologies. The goal of this study is to develop a hybrid algorithm for VLSI placement. The proposed algorithm includes a sequential combination of a genetic algorithm and an evolutionary algorithm. It is commonly known that local search algorithms, such as random forest, hill climbing, and variable neighborhoods, can be effectively applied to NP-hard problem-solving. They provide improved solutions, which are obtained after a global search. The scientific novelty of this research is based on the development of systems, principles, and methods for creating a hybrid (combined) placement algorithm. The principal difference in the proposed algorithm is that it obtains a set of alternative solutions in parallel and then selects the best one. Nonstandard genetic operators, based on problem knowledge, are used in the proposed algorithm. An investigational study shows an objective-function improvement of 13%. The time complexity of the hybrid placement algorithm is O(N²).     

Low‐temperature aging of baddeleyite‐based Ca‐TZP ceramics

The aging kinetics during low‐temperature aging of calcia‐stabilized tetragonal zirconia polycrystal (Ca‐TZP) ceramics prepared by high‐energy milling of natural zirconia mineral (baddeleyite) was studied by X‐ray diffraction under hydrothermal treatment conditions. Aging kinetics was investigated for ceramics with different contents of calcia. It was found that the kinetics may be well‐described within Johnson‐Mehl‐Avrami‐Kolmogorov model. Model parameters were determined by data fitting procedure. Change in exponential factor within Johnson‐Mehl‐Avrami‐Kolmogorov model with time is shown. Analytical model to describe aging kinetics is proposed. The transformation nucleation rate, initial diameter, and depth of the transformed areas and their growth rates are estimated. Degradation of hardness and fracture toughness is also reported for Ca‐TZP after low‐temperature aging for different contents of the stabilizer.     

Composition and surface energy characteristics of new petroleum resins

We investigated the composition and surface energy properties—the surface free energy, acid–base and dispersive components, acid and base parameters (according to the van Oss–Chaudhury–Good method) and acidity parameter (according to the Berger method) for several types of aromatic petroleum resins (PR). We could see that the Berger, nonlinear systems, and spatial methods provide slightly different information about the acid–base properties of PRs. For acid–base properties, relevant proton content was determined for each sample by nuclear magnetic resonance spectroscopy. Correlations between the composition and surface characteristics of the studied aromatic PR have been revealed. According to the data obtained from the acid–base approach and nuclear magnetic resonance spectroscopy, we can predict the possible combinations of PR–polymer with the best interface interaction, which can lead to high mechanical properties. POLYM. ENG. SCI., 57:1028–1032, 2017. © 2016 Society of Plastics Engineers     

Convenient approach to making nanocomposites based on a chitosan–poly(vinyl pyrrolidone) polymer matrix and a graphene nanofiller

Conducting, mechanically durable, elastic nanocomposite films were prepared with chitosan (CS) as the polymer matrix, graphene obtained from highly exfoliated graphite as the nanofiller, and poly(vinyl pyrrolidone) (PVP) as the stabilizer of the graphene sheets. The maximum graphene content in the composites without a loss of uniformity and other useful properties increased up to 4.0 wt %. The resulting composites were characterized by scanning electron microscopy, Raman spectroscopy, X‐ray diffraction analysis, mechanical testing, and electrical conductivity testing to determine the effects of the addition of graphene on the morphology and mechanical and electrical properties of the CS–PVP–graphene nanocomposite films. In this study, we took an approach to making nanocomposites from the perspectives of green chemistry, environmental protection, regenerative medicine, and low cost. © 2017 Wiley Periodicals, Inc. J. Appl. Polym. Sci. 2017 , 134, 45038.     

Multi-core implementation of decomposition-based packet classification algorithms

Multi-field packet classification is a network kernel function where packets are classified based on a set of predefined rules. Decomposition-based classification approaches are of major interest to the research community because of the parallel search in each packet header field. This paper presents four decomposition-based approaches on multi-core processors. We search in parallel for all the fields using linear search or range-tree search; we store the partial results in a linked list or a bit vector. The partial results are merged to produce the final packet header match. We evaluate the performance with respect to latency and throughput varying the rule set size (1–64 K). Experimental results show that our approaches can achieve 128 ns latency per packet and 11.5 Gbps throughput on state-of-the-art 16-core platforms.     

Functional synthesis for linear arithmetic and sets

Synthesis of program fragments from specifications can make programs easier to write and easier to reason about. To integrate synthesis into programming languages, synthesis algorithms should behave in a predictable way—they should succeed for a well-defined class of specifications. To guarantee correctness and applicability to software (and not just hardware), these algorithms should also support unbounded data types, such as numbers and data structures. To obtain appropriate synthesis algorithms, we propose to generalize decision procedures into predictable and complete synthesis procedures. Such procedures are guaranteed to find the code that satisfies the specification if such code exists. Moreover, we identify conditions under which synthesis will statically decide whether the solution is guaranteed to exist and whether it is unique. We demonstrate our approach by starting from a quantifier elimination decision procedure for Boolean algebra of set with Presburger arithmetic and transforming it into a synthesis procedure. Our procedure also works in the presence of parametric coefficients. We establish results on the size and the efficiency of the synthesized code. We show that such procedures are useful as a language extension with implicit value definitions, and we show how to extend a compiler to support such definitions. Our constructs provide the benefits of synthesis to programmers, without requiring them to learn new concepts, give up a deterministic execution model, or provide code skeletons.     

Distance-based tractography in high angular resolution diffusion MRI

High angular resolution diffusion imaging (HARDI) is a magnetic resonance imaging (MRI) technique, determining the diffusion of water molecules in tissue in vivo. HARDI is advantageous over the well-known diffusion tensor imaging (DTI), since it is able to extract more than one fiber orientation within a voxel and can therefore resolve crossing, kissing or fanning fiber tracts. However, multiple orientations per voxel require more sophisticated tractography approaches. In this paper we introduce a new deterministic fiber tracking method using the complete orientation distribution function (ODF) reconstructed from Q-ball imaging to enable tractography in challenging regions. Anisotropy classifiers are used to differentiate intra-voxel fiber populations and adjust a curvature threshold for one and multiple fiber configurations, respectively. In addition, we determine the most appropriate propagation direction in complex white matter regions, using the course of the current tract. To ensure tractography running within fiber bundles, a distance-based approach is integrated, which aims to maintain the initial distance of the seed point to the white matter boundary through the whole tracking. We evaluated our method using a phantom dataset featuring crossing, kissing and fanning fiber configurations and a human brain dataset, reconstructing the fanning of the corpus callosum and considering the region of the centrum semiovale.     

Imaging and Spectroscopic Sensing with Low-Repetition-Rate Terahertz Pulses and GaN TeraFET Detectors

Aiming for non-destructive testing and security applications, we investigate transmission-mode imaging and spectroscopic sensing using terahertz (THz) pulses from a commercial optical parametric oscillator (OPO) in combination with THz detectors based on antenna-coupled field-effect transistors (TeraFETs). The Q-switched OPO generates quasi-continuous-wave THz pulses with a peak power of up to 1 W at a repetition rate between 12 and 90 Hz. The pulses are frequency-tunable between 0.7 and 2.6 THz with a typical linewidth of 50 GHz. We explore detection with fast GaN/AlGaN TeraFETs which hold the potential for multi-pixel and homodyne detection.