Learning long-term dependencies with gradient descent is difficult

Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered.     

Leiomyomatosis peritonealis disseminata. Two case reports

AIM: To study the action of quercetin (Que) on inhibiting platelet aggregation. METHODS: Active oxygen free radicals produced by xanthine/xanthine oxidase (Xan/XO) reaction was used, platelet aggregation was determined by the turbidimetric method, and the Xan/XO oxyradicals generating reaction by luminol-dependent chemiluminescence (Che) method. RESULTS: Active oxygen free radicals enhanced the platelet aggregation induced by ADP 1.6 mumol.L-1. The rate of maximal aggregation increased from 29%-38% for ADP to 59%-70% for ADP + Xan/XO. The enhancement was abolished by the treatment of platelet-rich plasma (PRP) with Que 650 mumol.L-1 or hydrocortisone (Hyd) 900 mg.L-1. Both Que and Hyd scavenged the active oxyradicals in vitro. The Che was decreased by 75.7% (Que 4 mumol.L-1) and 79.0% (Hyd 900 mg.L-1) as compared with control. CONCLUSION: Active oxygen free radicals participated in the platelet aggregation, and scavenging oxyradicals by Que was one of mechanisms of inhibiting platelet aggregation.     

Introducing computer tools into a first course in electricalengineering

At a school where engineering students have immediate access to a personal computer, it is important that they become familiar with that tool early in their studies. In introductory engineering courses it is also important that the students learn to perceive the computer as a tool and not as an end in itself. Achieving these goals requires a careful integration of computer usage into such courses by the instructors. In this paper we examine this issue in the context of a first course in electrical engineering that has been taught at the US Naval Academy, and we present specific examples of assignments we have used in our effort to accomplish these goals     

DETERMINATION OF PALAEOSTRESSES FROM MESOSCALE SHEAR FRACTURES IN CORE SAMPLES USING THE MULTI-INVERSE METHOD

Palaeostress reconstruction is the key to quantitative characterization of fractured hydrocarbon reservoirs. In order to calculate the palaeostresses responsible for fault activity, faults in oriented core samples can be analysed and the optimal stresses can then be determined from fault-slip data in a least-squares sense. Many fractured reservoir rocks, however, include faults generated during a number of different tectonic phases. In these cases, the optimal stresses can be determined for subgroups of faults which are considered to be of approximately the same age. The difficulty with this is that criteria for accurate age-dating are often absent, especially from core samples.
Recently, the so-called "multi-inverse" method has been proposed to address this problem. This is a numerical technique for separating palaeostresses within a heterogeneous fault assemblage in the absence of a priori information on the faults'relative ages. In this paper, we apply the method to eighteen faults in two short (9m) cores from an exploration well in Hokkaido, northern Japan. The cores came from a fractured hydrocarbon reservoir and were oriented by correlation with borehole images. Core lithologies included massive Cretaceous basalts in which there was no layering which would show fault offsets; for these samples, we determined the sense of shear by studying asymmetric microstructures on the fault surfaces. The results show that the rocks have experienced six different stress regimes. In general, these are compatible with the tectonic history of the study area as inferred from the regional stratigraphy and from macroscale geological structures.
These results show that the multi-inverse method can be used to extract palaeostress data from cores, thereby providing information which will assist with the exploration of (and production from) fractured hydrocarbon reservoirs.     

New fault locating system for air-insulated substations usingoptical current detector

The authors present the basic philosophy, construction, design features, and test results of a new fault locating system. This fault locating system helps to shorten the time required for restoration of service after the occurrence of a busbar fault in an air-insulated distribution substation. Recent optical and electronic technologies allow the design of a highly accurate and compact fault locating system which consists of optical current detectors using the Faraday effect and a fault locating processor using a digital data processing technique. The fault location is carried out by discriminating between the directions of zero-sequence currents. Through various tests and field operations, it has been confirmed that the system has sufficient performance for practical applications     

Overgrowth and DNA synthesis of neuroepithelium in embryonic stages of induced Long-Evans rat myeloschisis

Overgrowth of the myeloschisis, namely the excessive amount of the neural plate tissue, has been reported in the human myeloschisis. However, it is still debatable how the overgrowth develops and whether the overgrowth is the cause, or the secondary effect of spinal dysraphism. The author induced myeloschisis in the fetuses of Long-Evans rats by the administration of ethylenethiourea (ETU) to pregnant rats on day 10 of gestation. The fetuses were removed 1 hour after the treatment with bromodeoxyuridine (BrdU) to the dams on day 14 and 21. The fetuses were fixed in alcohol and embedded in paraffin. H-E staining and the immunohistologic examination were performed on the staining patterns to anti-neurofilament (NFP), anti-glial fibrillary acidic protein (GFAP) and anti-BrdU antibody by ABC method. On day 14, the lateral portion of everted neural plate showed a loose arrangement of cells and there was rosette formation in the mesoderm. On day 21, cell necrosis was observed at the dorsolateral portion of myeloschisis, although the ventral portion showed almost normal cytoarchitecture and was positive to NFP and GFAP. The cause of myeloschisis in this model is supposed to be the local and direct cytotoxic effect of ETU to neuro-ectodermal junction. On day 14, control animals contained few BrdU-incorporated cells at the basal plate of neural tube. In contrast, everted neural plate showed an active uptake of BrdU diffusely in the subependymal matrix layer cells. Overgrowth was not yet identified. On day 21, overgrowth of myeloschisis was found in spite of a few positive cells to BrdU which was identical to the control animals. These findings seem to suggest that cells in the myeloschisis retain their ability of DNA synthesis for longer periods of development and overgrowth found on day 21 is possibly a secondary effect of spinal dysraphism in this model.     

Idiopathic epilepsy in the dog

The authors present a brief history of the development of the Ilizarov external fixator, and the classic uses associated with it. A preliminary study of 3 patients using 56 frames is presented. The Ilizarov external fixator is successfully used in these cases in place of traditional fixation.     

Dynamic programming implementation on array processor architectures

Dynamic Programming (DP) applies to many signal and image processing applications including boundary following, the Viterbi algorithm, dynamic time warping, etc. This paper presents an array processor implementation of generic dynamic programming. Our architecture is a SIMD array attached to a host computer. The processing element of the architecture is based on an ASIC design opting for maximum speed-up. By adopting a torus interconnection network, a dual buffer structure, and a multilevel pipeline, the performance of the DP chip is expected to reach the order of several GOPS. The paper discusses both the dedicated hardware design and the data flow control of the DP chip and the total array.This work was supported in part by the NATO, Scientific and Environmental Affairs Division, Collaborative Research Grant SA.5-2-05(CRG.960201)424/96/JARC-501.