Accelerating the convergence of the back-propagation method

The utility of the back-propagation method in establishing suitable weights in a distributed adaptive network has been demonstrated repeatedly. Unfortunately, in many applications, the number of iterations required before convergence can be large. Modifications to the back-propagation algorithm described by Rumelhart et al. (1986) can greatly accelerate convergence. The modifications consist of three changes:1) instead of updating the network weights after each pattern is presented to the network, the network is updated only after the entire repertoire of patterns to be learned has been presented to the network, at which time the algebraic sums of all the weight changes are applied:2) instead of keeping , the learning rate (i.e., the multiplier on the step size) constant, it is varied dynamically so that the algorithm utilizes a near-optimum , as determined by the local optimization topography; and3) the momentum factor is set to zero when, as signified by a failure of a step to reduce the total error, the information inherent in prior steps is more likely to be misleading than beneficial. Only after the network takes a useful step, i.e., one that reduces the total error, does again assume a non-zero value. Considering the selection of weights in neural nets as a problem in classical nonlinear optimization theory, the rationale for algorithms seeking only those weights that produce the globally minimum error is reviewed and rejected.     

Restriction fragment length polymorphisms of the phytohemagglutinin genes inPhaseolus andVigna (Leguminosae)

Restriction fragment length polymorphisms at the phytohemagglutinin (PHA) locus were determined among 21 genotypes ofPhaseolus vulgaris, P. coccineus, P. acutifolius, P. lunatus, and threeVigna species, using five restriction enzymes and one double digestion, in order to provide molecular evidence for their genetic relatedness. The dissimilarity between genotypes was estimated from binary RFLP data. The dissimilarity was high among species (from 0.75 to 0.95), and of variable extent among genotypes of the same species (0.33–0.89). InP. vulgaris, two different DNA hybridization patterns were found, giving further evidence for two major gene pools in that species. The restriction patterns ofP. vulgaris var.aborigineus, the putative ancestral form ofP. vulgaris, exhibit clear homology toP. vulgaris genotypes. An undefined landrace from Taiwan could be identified as aP. vulgaris genotype. RFLP-based trees for the phytohemagglutinin genes of the species studied were computed with several distance matrix and parsimony methods.     

Population genetics and phylogenetics of DNA sequence variation at multiple loci within the Drosophila melanogaster species complex

Two regions of the genome, a 1-kbp portion of the zeste locus and a 1.1- kbp portion of the yolk protein 2 locus, were sequenced in six individuals from each of four species: Drosophila melanogaster, D. simulans, D. mauritiana, and D. sechellia. The species and strains were the same as those of a previous study of a 1.9-kbp region of the period locus. No evidence was found for recent balancing or directional selection or for the accumulation of selected differences between species. Yolk protein 2 has a high level of amino acid replacement variation and a low level of synonymous variation, while zeste has the opposite pattern. This contrast is consistent with information on gene function and patterns of codon bias. Polymorphism levels are consistent with a ranking of effective population sizes, from low to high, in the following order: D. sechellia, D. melanogaster, D.mauritiana, and D. simulans. The apparent species relationships are very similar to those suggested by the period locus study. In particular, D. simulans appears to be a large population that is still segregating variation that arose before the separation of D. mauritiana and D. sechellia. It is estimated that the separation of ancestral D. melanogaster from the other species occurred 2.5-3.4 Mya. The separations of D. sechellia and D. mauritiana from ancestral D. simulans appear to have occurred 0.58- 0.86 Mya, with D. mauritiana having diverged from ancestral D. simulans 0.1 Myr more recently than D. sechellia.      

Further studies on non-H-2 histocompatibility alloantibodies in the mouse

Immunofluorescence tests indicate that alloantibodies specific for mouse histocompatibility antigens H-1^a, H-3^a, and H-13^a have been produced, using four different immunizations. Furthermore, an immunization employing donors and recipients which were H-2^k at the MHC produced stronger anti-H-3^a and anti-H-13^a than did immunizations where donors and recipients were H-2^b at the MHC.     

Classification of virulent and temperate bacteriophages of Listeria spp. on the basis of morphology and protein analysis.

A set of 22 phages of Listeria species (listeriaphages) (21 temperate and 1 virulent) were compared on the basis of morphology and protein composition. All 22 phages had icosahedral heads and exhibited either contractile or noncontractile tails. They represented two different morphotypes. Twenty phages belonged to the Siphoviridae family and could be differentiated only on the basis of tail length. Accordingly, they could be assigned to previously defined listeriaphage species. Two other phages were classified as members of the Myoviridae family, one of which (A511) should be regarded as a new species. It was found to be substantially different from all other known listeriaphages. All phages exhibited typical protein profiles, which were determined by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and subsequent laser densitometrical analysis of the gels. It was then possible to distinguish eight protein subgroups on the basis of unique protein patterns. This classification corresponds well to the previous groupings based on host range. Most of the phages revealed two or three major proteins ranging from 21 to 24 kDa and 30 to 36 kDa. In addition, at least 10 minor proteins could be observed for each phage. Our results indicate that the major proteins are structural proteins of the capsid and tail, and the protein band ranging from 30 to 35 kDa could clearly be assigned to the proteins of the phage capsid.