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.     

Calcium activates and inactivates a photoreceptor soma potassium current.

Light-induced currents were measured with a two-microelectrode voltage clamp of type B photoreceptor somata, which had been isolated by axotomy from all synaptic interactions as well as from all membranes capable of generating impulse activity. In artificial seawater (ASW), light elicited a transient early inward current, INa+, which depended on Na+o and had a linear current-voltage relation and an extrapolated reversal potential of 30-40 mV (absolute). In 0-Na+ ASW, light elicited a transient short-latency outward current that dependent on K+o, increased exponentially with more positive voltages (greater than or equal to -40 mV), and reversed at -70 to -75 mV. This outward current was not blocked by Ca++ channel blockers (e.g., Cd++, Co++) or substitution of Ba++o, for Ca++o, but was reduced by iontophoretic injection of EGTA. In both ASW and 0-Na+ ASW, light also elicited a delayed, apparently inward current, which was associated with a decreased conductance, depended on K+o, increased exponentially with more positive voltages (greater than or equal to -40 mV), reversed at the equilibrium potential for K+ flux in elevated K+o was eliminated by substitution of Ba++o for Ca++o, and was greatly reduced by Cd++o or Co++o. Thus, light elicited an early Ca++-dependent K+ current, IC, and a prolonged decrease of IC. Iontophoretic injection of Ca++ through a third microelectrode caused prolonged reduction of both IC and the light-induced decrease of IC, but did not alter ICa++ or the current-voltage relation of IC. Ruthenium red (1 microM) in the external medium caused a prolongation of the light-induced decrease of IC. Iontophoretic injection of EGTA often eliminated the light-induced IC decrease while decreasing peak IC (during depolarizing steps to -5 or 0 mV) by less than one-half. EGTA injection, on the average, did not affect steady state IC but reduced the light-induced decrease of steady state IC to approximately one-third of its original magnitude. The prolonged IC decrease, elicited by dim light in the absence of light-induced IC or INa+, was more completely eliminated by EGTA injection. It was concluded that light, in addition to inducing a transient inward Na+ current, causes both a transient increase and a prolonged decrease of IC via elevation of Ca++i.     

Reconstruction of ionic currents in a molluscan photoreceptor.

Two-microelectrode voltage-clamp measurements were made to determine the kinetics and voltage dependence of ionic currents across the soma membrane of the Hermissenda type B photoreceptor. The voltage-dependent outward potassium currents, IA and ICa(2+)-K+, the inward voltage-dependent calcium current, ICa2+ and the light-induced current, IIgt, were then described with Hodgkin-Huxley-type equations. The fast-activating and inactivating potassium current, IA, was described by the equation; IA(t) = gA(max)(ma infinity[1-exp(-t/tau ma)])3 x (ha infinity [1-exp(-t/tau ha)] + exp(-t/tau ha)) (Vm-EK), where the parameters ma infinity, ha infinity, tau ma, and tau ha are functions of membrane potential, Vm, and ma infinity and ha infinity are steady-state activation and inactivation parameters. Similarly, the calcium-dependent outward potassium current, ICa(2+)-K+, was described by the equation, ICa(2+)-K+ (t) = gc(max)(mc infinity(VC)(1-exp[-t/tau mc (VC)]))pc (hc infinity(VC) [1-exp(-t/tau hc)] + exp(-t/tau hc(VC)])pc(VC-EK). In high external potassium, ICa(2+)-K+ could be measured in approximate isolation from other currents as a voltage-dependent inward tail current following a depolarizing command pulse from a holding potential of -60 mV. A voltage-dependent inward calcium current across the type B soma membrane, ICa2+, activated rapidly, showed little inactivation, and was described by the equation: ICa2+ = gCa(max) [1 + exp](-Vm-5)/7]-1 (Vm-ECa), where gCa(max) was 0.5 microS. The light-induced current with both fast and slow phases was described by: IIgt(t) = IIgt1 + IIgt2 + IIgt3, IIgti = gIgti [1-exp(- ton/tau mi)] exp(-ton/tau hi)(Vm-EIgti) (i = 1, 2).(ABSTRACT TRUNCATED AT 250 WORDS)     

Different ionic conductances are modulated during the late receptor potential and the prolonged depolarizing afterpotential in Hermissenda type A photoreceptors

Wavelength-dependent, bistable phenomena were found in the receptor potential of Hermissenda crassicornis type A photoreceptors. Short exposure to blue light induced a prolonged depolarizing afterpotential (PDA) following the cessation of the light stimulus. Stronger adaptation to blue light, as caused by prolonged exposure and/or high intensity stimulation, effected a reduction in the early depolarizing transient of the late receptor potential (LRP) as elicited by subsequent stimuli. Vast separation of LRP emergence and PDA emergence could be obtained in photoreceptors in which a strong cancellation of the LRP was accomplished but a PDA still emerged after cessation of the light stimulus. Short exposure to yellow light cancelled the PDA, and stronger adaptation restored the LRP (opposite effect to blue light). The initial depolarizing part of the LRP had earlier been demonstrated to be mediated by the lightdependent increase of an inward conductance. In contrast, in this study the PDA was found to be accompanied by the reduction of an outward conductance, most likely a K⁺ conductance. A bistable photopigment system is thought to control the bistable receptor potential phenomenology by regulating the different membrane conductances during the LRP and the PDA.Abbreviations LRP late receptor potential - PDA prolonged depolarizing afterpotential - PHA prolonged hyperpolarizing afterpotential     

Interaction of chemosensory, visual, and statocyst pathways in Hermissenda crassicornis

Neurons in the cerebropleural ganglia (CPG), photoreceptors in the eye, optic ganglion cells, and statocyst hair cells of the nudibranch mollusk Hermissenda crassicornis responded in specific ways, as recorded intracellularly, to stimulation of the chemosensory pathway originating at the tentacular chemoreceptors as well as to stimulation of the visual pathway originating at the photoreceptors. Synaptic inhibition of photoreceptors occurs via the chemosensory pathway. The possible significance of such intersensory interaction is discussed with reference to preliminary investigation of the animal's gustatory behavior and possible neural mechanisms of behavioral choice.     

Calcium-mediated decrease of a voltage-dependent potassium current.

Elevated intracellular Ca++ concentration reduces the amplitude of an early, voltage-dependent K+ current (IA) in the Type B photoreceptor of Hermissenda crassicornis. Internal Ca++ is increased by activating a voltage and light-dependent Ca++ current present in these cells or by direct iontophoresis of Ca++ ions. Substitution of Ba++ for Ca++ or elimination of Ca++ from the sea water bathing the cells abolishes the reduction in IA during paired light and depolarizing voltage steps. The delayed K+ current (IB) in these cells is also reduced during paired light and voltage steps, but this decrease of IB is not affected by removal of extracellular Ca++. IB (but not IA), apparently much less dependent on intracellular Ca++ levels, is reduced by light alone. Ca++ iontophoresis also abolishes the light-dependent Na+ current, which recovers with a time course of minutes.     

Classical Conditioning-Induced Changes in Low-Molecular-Weight GTP-Binding Proteins in Rabbit Hippocampus

Classical conditioning of Hermissenda, involving paired light-rotation events, results in a 30-35% decrease in the levels of a 20-kDa G protein (cp20). To test whether a similar protein exists in vertebrates, rabbits were trained to associate a tone with periorbital electrical stimulation and G proteins were analyzed by photoaffinity labeling with [alpha-32P]GTP-azidoanilide. A 20-kDa G protein similar to cp20 decreased by 36% in the hippocampus of rabbits subjected to paired tone and electrical stimulation, but not in unpaired controls. Learning-specific decreases were also found in the amount of ras protein.     

Age-Specific Mortality During the 1918 Influenza Pandemic: Unravelling the Mystery of High Young Adult Mortality

The worldwide spread of a novel influenza A (H1N1) virus in 2009 showed that influenza remains a significant health threat, even for individuals in the prime of life. This paper focuses on the unusually high young adult mortality observed during the Spanish flu pandemic of 1918. Using historical records from Canada and the U.S., we report a peak of mortality at the exact age of 28 during the pandemic and argue that this increased mortality resulted from an early life exposure to influenza during the previous Russian flu pandemic of 1889–90. We posit that in specific instances, development of immunological memory to an influenza virus strain in early life may lead to a dysregulated immune response to antigenically novel strains encountered in later life, thereby increasing the risk of death. Exposure during critical periods of development could also create holes in the T cell repertoire and impair fetal maturation in general, thereby increasing mortality from infectious diseases later in life. Knowledge of the age-pattern of susceptibility to mortality from influenza could improve crisis management during future influenza pandemics.

“The war is over – and I must go” Egon Schiele, 1890–1918.

     

Sequential modification of membrane currents with classical conditioning

Pavlovian conditioning of the nudibranch mollusc Hermissenda crassicornis was previously shown to produce long-lasting reduction of two K+ currents measured across the Type B photoreceptor soma membrane (Alkon et al., 1982a; Alkon et al., 1985). Pavlovian conditioning of the rabbit was also shown to be followed by persistent K+ current reduction (Disterhoft et al., 1986). Here we report the first evidence that Ca2+ currents can also be modified by conditioning. The amplitude of the currents rather than their voltage-dependence remains reduced at least 1-2 d after conditioning (but not control procedures). Conditioning-induced changes of both K+ and Ca2+ currents increased as a function of training, the Ca2+ currents only changing substantially with greater than or equal to 250 trials. The later changes of the Ca2+ current may function to limit the magnitude of excitability increases due to associative learning.     

Calcium-regulated GTPase activity in the calcium-binding protein calexcitin

Calexcitin (CE) is a calcium-binding protein, closely related to sarcoplasmic calcium-binding proteins, that is involved in invertebrate learning and memory. Early reports indicated that both Hermissenda and squid CE also could bind GTP; however, the biochemical significance of GTP-binding and its relationship to calcium binding have remained unclear. Here, we report that the GTPase activity of CE is strongly regulated by calcium. CE possessed a P-loop-like structure near the C-terminal similar to the phosphate-binding regions in other GTP-binding proteins. Site-directed mutagenesis of this region showed that Gly¹⁸², Phe¹⁸⁶ and Gly¹⁸⁷ are required for maximum affinity, suggesting that the GTP-binding motif is G-N-x-x-[FM]-G. CE cloned from Drosophila CNS possessed a similar C-terminal sequence and also bound and hydrolyzed GTP. GTPase activity in Drosophila CE was also strongly regulated by Ca²⁺, exhibiting over 23-fold higher activity in the presence of 0.3 μM calcium. Analysis of the conserved protein motifs defines a new family of Ca²⁺-binding proteins representing the first example of proteins endowed with both EF-hand calcium binding domains and a C-terminal, P-loop-like GTP-binding motif. These results establish that, in the absence of calcium, both squid and Drosophila CE bind GTP at near-physiological concentrations and hydrolyze GTP at rates comparable to unactivated ras. Calcium functions to increase GTP-binding and GTPase activity in CE, similar to the effect of GTPase activating proteins in other low-MW GTP-binding proteins. CE may, therefore, act as a molecular interface between Ca²⁺ cytosolic oscillations and the G protein-coupled signal transduction.