Performance evaluation of neural network architectures: the case of predicting foreign exchange correlations

In the last decade, neural networks have emerged from an esoteric instrument in academic research to a rather common tool assisting auditors, investors, portfolio managers and investment advisors in making critical financial decisions. It is apparent that a better understanding of the network's performance and limitations would help both researchers and practitioners in analysing real‐world problems. Unlike many existing studies which focus on a single type of network architecture, this study evaluates and compares the performance of models based on two competing neural network architectures, the multi‐layered feedforward neural network (MLFN) and general regression neural network (GRNN). Our empirical evaluation measures the network models' strength on the prediction of currency exchange correlation with respect to a variety of statistical tests including RMSE, MAE, U statistic, Theil's decomposition test, Henriksson–Merton market timing test and Fair–Shiller informational content test. Results of experiments suggest that the selection of proper architectural design may contribute directly to the success in neural network forecasting. In addition, market timing tests indicate that both MLFN and GRNN models have economically significant values in predicting the exchange rate correlation. On the other hand, informational content tests discover that the neural network models based on different architectures capture useful information not found in each other and the information sets captured by the two network designs are independent of one another. An auxiliary experiment is developed and confirms the possible synergetic effect from combining forecasts made by the two different network architectures and from incorporating information from an implied correlation model into the neural network forecasts. Implied correlation and random walk models are also included in our empirical experiment for benchmark comparison. Copyright © 2005 John Wiley & Sons, Ltd.     

Heterodimeric JAK-STAT activation as a mechanism of persistence to JAK2 inhibitor therapy

The identification of somatic activating mutations in JAK2 (refs?1–4) and in the thrombopoietin receptor gene (MPL) in most patients with myeloproliferative neoplasm (MPN) led to the clinical development of JAK2 kinase inhibitors. JAK2 inhibitor therapy improves MPN-associated splenomegaly and systemic symptoms but does not significantly decrease or eliminate the MPN clone in most patients with MPN. We therefore sought to characterize mechanisms by which MPN cells persist despite chronic inhibition of JAK2. Here we show that JAK2 inhibitor persistence is associated with reactivation of JAK–STAT signalling and with heterodimerization between activated JAK2 and JAK1 or TYK2, consistent with activation of JAK2 in trans by other JAK kinases. Further, this phenomenon is reversible: JAK2 inhibitor withdrawal is associated with resensitization to JAK2 kinase inhibitors and with reversible changes in JAK2 expression. We saw increased JAK2 heterodimerization and sustained JAK2 activation in cell lines, in murine models and in patients treated with JAK2 inhibitors. RNA interference and pharmacological studies show that JAK2-inhibitor-persistent cells remain dependent on JAK2 protein expression. Consequently, therapies that result in JAK2 degradation retain efficacy in persistent cells and may provide additional benefit to patients with JAK2-dependent malignancies treated with JAK2 inhibitors.     

High-throughput decoding of antitrypanosomal drug efficacy and resistance

The concept of disease-specific chemotherapy was developed a century ago. Dyes and arsenical compounds that displayed selectivity against trypanosomes were central to this work, and the drugs that emerged remain in use for treating human African trypanosomiasis (HAT). The importance of understanding the mechanisms underlying selective drug action and resistance for the development of improved HAT therapies has been recognized, but these mechanisms have remained largely unknown. Here we use all five current HAT drugs for genome-scale RNA interference target sequencing (RIT-seq) screens in Trypanosoma brucei, revealing the transporters, organelles, enzymes and metabolic pathways that function to facilitate antitrypanosomal drug action. RIT-seq profiling identifies both known drug importers and the only known pro-drug activator, and links more than fifty additional genes to drug action. A bloodstream stage-specific invariant surface glycoprotein (ISG75) family mediates suramin uptake, and the AP1 adaptin complex, lysosomal proteases and major lysosomal transmembrane protein, as well as spermidine and N-acetylglucosamine biosynthesis, all contribute to suramin action. Further screens link ubiquinone availability to nitro-drug action, plasma membrane P-type H(+)-ATPases to pentamidine action, and trypanothione and several putative kinases to melarsoprol action. We also demonstrate a major role for aquaglyceroporins in pentamidine and melarsoprol cross-resistance. These advances in our understanding of mechanisms of antitrypanosomal drug efficacy and resistance will aid the rational design of new therapies and help to combat drug resistance, and provide unprecedented molecular insight into the mode of action of antitrypanosomal drugs.     

The Allocation of Production Orders to Offshore Plants in the Clothing Industry:Part1-Model Conceptualisation

As a rough cut planning exercise,production order allo-cation for offshore manu facturing sites in the clothingindustry has received little attention as does other typesof planning like production planning.This paper outlinesan investigation into the exercise of production order al-location in the clothing industry.We have identified thatpractitioners involving in this exercise are not unanimousin the selection of offshore sites.Part 1 presents thebackground,objectives,process and conceptual model ofproduction order allocation.     

一种基于奇异值分解的带噪语音识别方法

提出了一种抗噪声的语音识别方法，用于训练和用于测试的语音信号在提取特征之前，均需经过相同的奇异值分解滤波，本文还提出了一种滤波参数的选取方法，实验证明，采用这种方法可以大幅度提高传统隐马尔可夫模型语音识别系统的抗噪声性能。     

Mutant small heat-shock protein 27 causes axonal Charcot-Marie-Tooth disease and distal hereditary motor neuropathy

Charcot-Marie-Tooth disease (CMT) is the most common inherited neuromuscular disease and is characterized by considerable clinical and genetic heterogeneity. We previously reported a Russian family with autosomal dominant axonal CMT and assigned the locus underlying the disease (CMT2F; OMIM 606595) to chromosome 7q11-q21 (ref. 2). Here we report a missense mutation in the gene encoding 27-kDa small heat-shock protein B1 (HSPB1, also called HSP27) that segregates in the family with CMT2F. Screening for mutations in HSPB1 in 301 individuals with CMT and 115 individuals with distal hereditary motor neuropathies (distal HMNs) confirmed the previously observed mutation and identified four additional missense mutations. We observed the additional HSPB1 mutations in four families with distal HMN and in one individual with CMT neuropathy. Four mutations are located in the Hsp20-alpha-crystallin domain, and one mutation is in the C-terminal part of the HSP27 protein. Neuronal cells transfected with mutated HSPB1 were less viable than cells expressing the wild-type protein. Cotransfection of neurofilament light chain (NEFL) and mutant HSPB1 resulted in altered neurofilament assembly in cells devoid of cytoplasmic intermediate filaments.