Multi-stage Computation Research Articles

Normalization by orientation-tuned surround in human V1-V3.

An influential account of neuronal responses in primary visual cortex is the normalized energy model. This model is often implemented as a multi-stage computation. The first stage is linear filtering. The second stage is the extraction of contrast energy, whereby a complex cell computes the squared and summed outputs of a pair of the linear filters in quadrature phase. The third stage is normalization, in which a local population of complex cells mutually inhibit one another. Because the population includes cells tuned to a range of orientations and spatial frequencies, the result is that the responses are effectively normalized by the local stimulus contrast. Here, using evidence from human functional MRI, we show that the classical model fails to account for the relative responses to two classes of stimuli: straight, parallel, band-passed contours (gratings), and curved, band-passed contours (snakes). The snakes elicit fMRI responses that are about twice as large as the gratings, yet a traditional divisive normalization model predicts responses that are about the same. Motivated by these observations and others from the literature, we implement a divisive normalization model in which cells matched in orientation tuning ("tuned normalization") preferentially inhibit each other. We first show that this model accounts for differential responses to these two classes of stimuli. We then show that the model successfully generalizes to other band-pass textures, both in V1 and in extrastriate cortex (V2 and V3). We conclude that even in primary visual cortex, complex features of images such as the degree of heterogeneity, can have large effects on neural responses.

Real-time structural health monitoring system based on streaming data

In this paper, a novel real-time structural health monitoring (SHM) system based on streaming data is proposed. In contrast to a traditional SHM system, the proposed system implements a series of optimizations for data transmission and processing to reduce the latency and better satisfy the real-time requirement. The concept of the watermark in the streaming system is adopted to address the problem of when to trigger the time window calculation under the real-time requirement. Moreover, a well-designed parallel mechanism is used to satisfy the multistage computation requirement in the parallel data stream. A case study in which the proposed system is applied to the Shanghai Tower is presented. The peak picking method is used as an example in the test environment to track the latency of each main operation under different parallelism schemes. The results show that computing in parallel effectively reduces the latency and provides a reference for integrating the random decrement technique (RDT), stochastic subspace identification (SSI), or other more complex but effective algorithms in parallel into the system in the future. The total latency under the test environment from data generation to data transmission to the web server is approximately only 200-400 ms, which indicates the excellent real-time performance of the system.

Self-Spectre, Write-Execute and the Hidden State

Abstract The recent Meltdown and Spectre vulnerabilities have highlighted a very present and real threat in the on-chip memory cache units which can ultimately provide a hidden state, albeit only readable via memory timing instructions [Kocher, P.—Genkin, D.— Gruss, D.— Haas, W.—Hamburg, M.—Lipp, M.–Mangard, S.—Prescher, T.—Schwarz, M.—Yarom, Y.: Spectre attacks: Exploiting speculative execution, CoRR, abs/1801.01203, 2018]. Yet the exploits, although having some complexity and slowness, are demonstrably reliable on nearly all processors produced for the last two decades. Moving out from looking at this strictly as a means of reading protected memory, as the large microprocessor companies move to close this security vulnerability, an interesting question arises. Could the inherent design of the processor give the ability to hide arbitrary calculations in this speculative and parallel side channel? Without even using protected memory and exploiting the vulnerability, as has been the focus, there could very well be a whole class of techniques which exploit the side-channel. It could be done in a way which would be largely un-preventable behavior as the technology would start to become self-defeating or require a more complicated and expensive on-chip cache memory system to properly post-speculatively clean itself. And the ability to train the branch predictor to incorrectly speculatively behave is almost certain given hardware limitations, andthusprovidesexactly this pathway. A novel approach looks at just how much computation can be done speculatively with a result store via indirect reads and available through the memory cache. A multi-threaded approach can allow a multi-stage computation pipeline where each computation is passed to a read-out thread and then to the next computation thread [Swanson, S.—McDowell, L. K.—Swift, M. M.—Eggers, S. J.–Levy H. M.: An evaluation of speculative instruction execution on simultaneous multithreaded processors, ACM Trans. Comput. Syst. 21 (2003), 314–340]. Through channels like this, an application can surreptitiously make arbitrary calculations, or even leak data without any standard tracing tools being capable of monitoring the subtle changes. Like a variation of the famous physics Heisenberg uncertainty principle, even a tool capable of reading the cache states would not only be incredibly inefficient, but thereby tamper with and modify the state. Tools like in-circuit emulators, or specially designed cache emulators would be needed to unmask the speculative reads, and it is further difficult to visualize with a linear time-line. Specifically, the AES and RSA algorithms will be studied with respect to these ideas, looking at success rates for various calculation batches with speculative execution, while having a summary view to see the rather severe performance penalties for using such methods. Either approaches could provide for strong white-box cryptography when considering a binary, non-source code form. In terms of white-box methods, both could be significantly challenging to locate or deduce the inner workings of the code. Further, both methods can easily surreptitiously leak or hide data within shared memory in a seemingly innocuous manner.

Application of the multicriterion optimization techniques and hierarchy of computational models to the research of ion acceleration due to laser-plasma interaction

In this paper we discuss how a particles-in-cell computation code can be combined with methods of multicriterion optimization (in particular the Pareto optimal solutions of the multicriterion optimization problem) and a hierarchy of computational models approach to create an efficient tool for solving a wide array of problems related to the laser-plasma interaction. In case of the computational experiment the multicriterion optimization can be applied as follows: the researcher defines the objectives of the experiment - some computable scalar values (i.e. high kinetic energy of the ions leaving the domain, least possible number of electrons leaving domain in the given direction, etc). After that the parameters of the experiment which can be varied to achieve these objectives and the constrains on these parameters are chosen (e.g. amplitude and wave-length of the laser radiation, dimensions of the plasma slab(s)). The Pareto optimality of the vector of the parameters can be seen as this: x0 is Pareto optimal if there exists no vector which would improve some criterion without causing a simultaneous degradation in at least one other criterion. These efficient set of parameter and constrains can be selected based on the preliminary calculations in the simplified models (one or two-dimensional) either analytical or numerical. The multistage computation of the Pareto set radically reduces the number of variants which are to be evaluated to achieve the given accuracy. During the final stage we further improve the results by recomputing some of the optimal variants on the finer grids, with more particles and/or in the frame of a more detailed model. As an example we have considered the ion acceleration caused by interaction of very intense and ultra-short laser pulses with plasmas and have calculated the optimal set of experiment parameters for optimizing number and average energy of high energy ions leaving the domain in the given direction and minimizing the expulsion of electrons.

Simultaneous Interpolation and Approximation by a Class of Multivariate Positive Operators

The purpose of this paper is to show that the interpolation positive operators of a wide class satisfy also the approximation property. Such a situation of simultaneous interpolation and approximation may be very desirable, but is rather unusual. Our attention is focused on the convergence problem, giving the conditions under which a sequence of operators of the considered class converges to a continuous function in a convex compact set in Rm (m∈N). It must be recalled that many of these operators are very interesting in applications and that suitable algorithms can be devised for parallel, multistage and iterative computation.

A sound reduction semantics for untyped CBN multi-stage computation. Or, the theory of MetaML is non-trival

A multi-stage computation is one involving more than one stage of execution. MetaML is a language for programming multi-stage computations. Previous studies presented big-step semantics, categorical semantics, and sound type systems for MetaML. In this paper, we report on a confluent and sound reduction semantics for untyped call-by name (CBN) MetaML. The reduction semantics can be used to formally justify some optimization performed by a CBN MetaML implementation. The reduction semantics demonstrates that non-trivial equalities hold for object-code, even in the untyped setting. The paper also emphasizes that adding intensional analysis (that is, taking-apart object programs) to MetaML remains an interesting open problem.

Experimental and theoretical studies of countercurrent solvent-gas extraction—II: The construction and operation of a laboratory differential vapour-absorption column

The construction and operation of a simple 30 mm differential vapour-absorption column are described, the column being packed with a knitted wire packing. The results of separations of binary volatile mixtures are reported and analysed by means of a multistage computation procedure. It has been confirmed that in accordance with theoretical prediction, the separation efficiency can be improved in some cases by imposing longitudinal temperature differences in the column. Computed values of HETP are reported which show a linear increase with solvent rate in the range 0·2-1·01/hr. For the same loadings, HETP values are 50–300 per cent greater than for continuous gas chromatography and are about four times greater than those measured for a normal distillation process using the same packing material.

Experimental and theoretical studies of countercurrent solvent-gas extraction—I Continuous gas-chromatography in a laboratory moving-bed column

The experimental results of separation of binary volatile mixtures in a 25 mm column have been compared with the results of a multi-stage computation procedure based on established phase-equilibrium data. Very approximate agreement between experiment and theory is obtained assuming an HETP of 13–18 mm for the conditions used. The theoretical prediction of an optimum temperature has been qualitatively confirmed.