Complete and partial fault tolerance of feedforward neural nets

A method is proposed to estimate the fault tolerance (FT) of feedforward artificial neural nets (ANNs) and synthesize robust nets. The fault model abstracts a variety of failure modes for permanent stuck-at type faults. A procedure is developed to build FT ANNs by replicating the hidden units. It exploits the intrinsic weighted summation operation performed by the processing units to overcome faults. Metrics are devised to quantify the FT as a function of redundancy. A lower bound on the redundancy required to tolerate all possible single faults is analytically derived. Less than triple modular redundancy (TMR) cannot provide complete FT for all possible single faults. The actual redundancy needed to synthesize a completely FT net is specific to the problem at hand and is usually much higher than that dictated by the general lower bound. The conventional TMR scheme of triplication and majority voting is the best way to achieve complete FT in most ANNs. Although the redundancy needed for complete FT is substantial, the ANNs exhibit good partial FT to begin with and degrade gracefully. The first replication yields maximum enhancement in partial FT compared with later successive replications. For large nets, exhaustive testing of all possible single faults is prohibitive, so the strategy of randomly testing a small fraction of the total number of links is adopted. It yields partial FT estimates that are very close to those obtained by exhaustive testing. When the fraction of links tested is held fixed, the accuracy of the estimate generated by random testing is seen to improve as the net size grows.     

Signal pathway regulation of interleukin-8-induced actin polymerization in neutrophils

Interleukin-8 (IL-8), a recently described peptide cytokine, is a neutrophil chemoattractant and activator that exerts effects similar to fMLP, yet their receptors and their roles in pathophysiology differ. The effect of IL-8 on the neutrophil cytoskeleton has not been well studied; therefore, we compared and contrasted the effects of IL-8 and fMLP on neutrophil actin conformation and on the signal pathway regulation of actin responses. IL-8 caused a rapid, dose-dependent increase in neutrophil F-actin content within 30 seconds. The maximum increase was twofold. These changes were accompanied by the development of F-actin-rich pseudopods, as noted with fluorescence microscopy and scanning electron microscopy. Selected biochemical inhibitors were used to study the regulation of the IL-8-induced actin changes. Incubation of neutrophils with 2 micrograms/mL pertussis toxin resulted in a 67% inhibition of the IL-8-induced F-actin increase. The protein kinase C (PKC) inhibitors, staurosporine and H7, did not inhibit the increase in F-actin caused by IL-8. IL-8 caused a rapid increase in neutrophil intracellular calcium that could be completely inhibited by the chelating agent 1,2-bis(o-aminophenoxy)ethane-N,N-N',N'-tetraacetic acid (BAPTA). However, BAPTA-treated neutrophils retained the ability to increase F-actin in response to IL-8. Similar results were seen with fMLP, indicating that, similar to fMLP, the IL-8-induced actin response is mediated through pertussis-toxin-sensitive G-proteins but is neither dependent on PKC nor increases in cytosolic calcium. Thus, although IL-8 and fMLP exert their effects on neutrophils through different receptors, the signal transduction pathways used and the effects on actin conformation and pseudopod formation are similar.     

In-house development of co-fireable thick film silver conductor for LTCC applications

The rapidly growing wireless industry needs new high performance materials to build low loss, high density, and thermally stable integrated packages. Applications include automotive safety, control, global positioning system (GPS) mapping and entertainment, multifunctional portable phones, video and data transmission through wireless local area network (WLAN) etc. Recently, low temperature co-fired ceramics (LTCC) technology is referred as a key approach for smart packaging. Although its use has been initiated in the telecommunication field due to the excellent dielectric properties of the LTCC tapes, its application areas have been diversified recently. In the present work, the attempt has been made towards the development of thick film silver conductor composition compatible with available LTCC tapes (DuPont DP-951AX). The physical, microstructural, and electrical properties of the pastes prepared with different compositions are presented in this paper.     

Double insemination and gonadotropin-releasing hormone treatment of repeat-breeding dairy cattle

Our objective was to determine if double inseminations during the same estrous period of dairy cattle eligible for their third or fourth service (repeat breeders) would improve pregnancy rates equivalent to injections of GnRH given at the time of AI. Repeat-breeding, lactating cows from six herds (five herds in the San Joaquin Valley of central California and one herd in northeast Kansas) were assigned randomly to four treatment groups when detected in estrus: 1) single AI plus no injection, 2) single AI plus 100 micrograms GnRH at AI, 3) double AI plus no injection, or 4) double AI plus 100 micrograms of GnRH at AI. Inseminations were performed according to the a.m.-p.m. rule. The second AI for the double AI treatment was given 12 to 16 h after the first AI. Injections of GnRH were given intramuscularly immediately following the single AI or the first AI of the double AI. Pregnancy rates of cows given a single AI and hormone injection were numerically higher in all six herds than those of their herdmates given only a single AI. In five of six herds, the pregnancy rates of cows given a double AI and hormone injection were numerically higher than pregnancy rates of their herdmates given only a double AI. Overall pregnancy rates for the four treatments were 1) 112/353 (32.1%), 2) 165/406 (41.6%), 3) 119/364 (33.5%), and 4) 135/359 (37.5%). Gonadotropin-releasing hormone increased pregnancy rates of repeat breeders compared with controls given only a single AI. No further benefit beyond the single AI was accrued from the double AI treatment, with or without concurrent hormone administration.     

Solution route synthesis and characterization of nanocrystalline Bi<Subscript>2</Subscript>Ru<Subscript>2</Subscript>O<Subscript>7</Subscript> for usage in resistor paste application

In this paper, we report synthesis of Bi₂Ru₂O₇ by solution route following the usual procedure, wherein, RuCl₃ solution was added drop by drop to a solution of bismuth nitrate while maintaining the pH of reaction at 6.75 with the help of dilute HCl and NaOH. Various physico-chemico-thermal analysis of these as -precipitated powders reveal that Bi & Ru precipitate as complex hydroxides with particle size ranging between 100–200 nm. The XRD analyses of the vacuum-dried (250°C) samples indicate conversion to Bi₂Ru₂O₇ (BRO) phase with low crystallinity having some RuO₂ impurity. Further heat treatment at 600°C for 4 h in an air atmosphere produces highly crystalline Bi₂Ru₂O₇ powder with a trace impurity of RuO₂. Resistor pastes were made using these BRO powders by intimately mixing them with other ingredients such as standard lead alumino-boro-silicate glass frit and polymeric thixotropic agents like ethyl cellulose and organic solvents like butyl cellosolve and -Terpinol. For all the paste samples, the inorganic part of the paste contained 43% glass and 57% BRO. These pastes were screen-printed and fired onto alumina substrates with pre-fired in-house Pd-Ag conductor terminations. The fired resistors had sheet resistance of 35–75 and hot TCR between 150 to 480 ppm/°C and zero cold TCR for resistors No.10 & 18 and for other resistors the cold TCR deviated from zero due to low scale processing of resistor paste.     

Perfect fault tolerance of the n-k-n network

It was shown in Phatak, Choi, and Koren (1993) that the neural network implementation of the n-k-n encoder-decoder has a minimal size of k = 2, and it was shown how to construct the network. A proof was given in Phatak and Koren (1995) that in order to achieve perfect fault tolerance by replicating the hidden layer, the required number of replications is at least three. In this note, exact lower bounds are derived for the number of replications as a function of n, for the n-2-n network and for the n- log2n-n network.     

Synthesis of single phase Tl-2223 superconductors: How much thallium do we really need?

This paper describes the synthesis and microwave absorption studies of single phase Tl-2223 compound, withT _c=(129 ± 1) K. The single phase compound has been synthesized using a precursor matrix method. It has been found that Tl-deficient starting composition gives the best results.     

Generalized haar DWT and transformations between decision trees and neural networks

The core contribution of this paper is a three-fold improvement of the Haar discrete wavelet transform (DWT). It is modified to efficiently transform a multiclass- (rather than numerical-) valued function over a multidimensional (rather than low dimensional) domain, or transform a multiclass-valued decision tree into another useful representation. We prove that this multidimensional, multiclass DWT uses dynamic programming to minimize (within its framework) the number of nontrivial wavelet coefficients needed to summarize a training set or decision tree. It is a spatially localized algorithm that takes linear time in the number of training samples, after a sort. Convergence of the DWT to benchmark training sets seems to degrade with rising dimension in this test of high dimensional wavelets, which have been seen as difficult to implement. This multiclass multidimensional DWT has tightly coupled applications from learning "dyadic" decision trees directly from training data, rebalancing or converting preexisting decision trees to fixed depth boolean or threshold neural networks (in effect parallelizing the evaluation of the trees), or learning rule/exception sets represented as a new form of tree called an "E-tree," which could greatly help interpretation/visualization of a dataset.     

Inseminations at estrus induced by presynchronization before application of synchronized estrus and ovulation

A controlled field study examined conception rates after 2 timed artificial insemination (TAI) breeding protocols conducted on 2 commercial dairy farms. Estrous cycles in postpartum lactating cows were presynchronized with 2 injections of PGF(2alpha) given 14 d apart (Pre-synch) and then, after 12 d, the standard Ovsynch protocol (injection of GnRH 7 d before and 48 h after an injection of PGF(2alpha), with one TAI at 12 to 16 h after the second GnRH injection) or Heatsynch protocol [injection of GnRH 7 d before an injection of PGF(2alpha), followed 24 h later by 1 mg of estradiol cypionate (ECP) and one TAI 48 h after ECP] was applied. Experimental design allowed artificial insemination to occur anytime after the second Presynch injection and during the designed breeding week when estrus was detected. Of the 1846 first services performed, only 1503 (rate of compliance = 81.4%) were performed according to protocol. Numbers of cows inseminated, logistic-regression adjusted conception rates, and days in milk (DIM) were for inseminations made: 1) during 14 d after first Presynch injection (n = 145; 22.6%; 54 +/- 0.4 DIM); 2) during 12 d after second Presynch injection (n = 727; 33%; 59 +/- 0.2 DIM); 3) during 7 d after the first GnRH injection of Ovsynch or Heatsynch (n = 96; 32.1%; 74 +/- 0.5 DIM); 4) after estrus as part of Heatsynch (n = 212; 44.6%; 76 +/- 0.3 DIM); 4) after TAI as part of Heatsynch (n = 154; 21.1%; 76 +/- 0.4 DIM); 5) after estrus as part of Ovsynch (n = 43; 48.7%; 77 +/- 0.7 DIM); and 6) after TAI as part of Ovsynch (n = 271; 24.4%; 77 +/- 0.3 DIM). Conception rates when AI occurred after one Presynch injection were less than when AI occurred after 2 Presynch injections. Conception rates for those inseminated after either Presynch injection did not differ from those inseminated after combined Heatsynch + Ovsynch. Cows in the Ovsynch and Heatsynch protocols inseminated after estrus during the breeding week had greater conception rates than those receiving the TAI, but overall conception rates did not differ between protocols. Among cows inseminated after detected estrus, conception was greater for cows in the Heatsynch + Ovsynch protocol (77 +/- 0.4 DIM) than for those inseminated after either Presynch injection (54 +/- 0.4 or 59 +/- 0.2 DIM). We concluded that conception rates after Heatsynch and Ovsynch were similar under these experimental conditions, and that delaying first AI improved fertility for cows inseminated after detected estrus.