Search

This work describes an approach for inferring Deterministic Context-free (DCF) Grammars in a C!onnectionist pjiradigm using a Recurrent Neural Network Pushdown Automaton (NNPDA). The N N P D A consists of a recurrent neural network connected to an external stack memory through a common error function. W e show that the N N P D A is able to learn the dynamics of an underlying pushdown automaton from examples of grammatical and non-grammatical strings. Not only does the network learn the state transitions in the automaton, it also learns the actions required to control the stack. In order to use continuous optimization methods, we develop an analog stack which reverts to a discrete stack by quantization of all activations, after the network has learned the transition rules and stack actions. W e further show an enhancement of the network's learning capabilities by providing hints. In addition, an initial comparative study of simulations with first, second and third order recurrent networks has shown that the increased degree of freedom in a higher order networks improve generalization bu not necessarily learning speed.

It is often difficult to predict the optimal neural network size for a particular application. Constructive or destructive methods that add or subtract neurons, layers, connections, etc. might offer a solution to this problem. We prove that one method, recurrent cascade correlation, due to its topology, has fundamental limitations in representation and thus in its learning capabilities. It cannot represent with monotone (i.e., sigmoid) and hard-threshold activation functions certain finite state automata. We give a 'preliminary' approach on how to get around these limitations by devising a simple constructive training method that adds neurons during training while still preserving the powerful fully-recurrent structure. We illustrate this approach by simulations which learn many examples of regular grammars that the recurrent cascade correlation method is unable to learn.

This paper reports mineral and whole‐rock geochemical data (major and trace elements, Sr–Nd isotopes, and in situ plagioclase Sr‐isotope analysis) for Cenozoic volcanic rocks from Sabeto, NW Fiji, with the aim of constraining the origin of Cenozoic shoshonitic rocks found throughout Fiji. The most mafic shoshonitic lavas (absarokites) exhibit porphyritic textures and massive structure with phenocrysts of olivine (Fo = 76–78), clinopyroxene (Mg# = 73.9–80.0), and plagioclase (An = 71.9–92.1). Major element geochemical data include SiO2 = 47.31–48.46 wt%, MgO = 4.20–5.76 wt%, Mg# = 49–51, Al2O3 = 17.64–18.93 wt%, and (Na2O + K2O) = 5.12–7.01 wt%. The trace element geochemistry is typically arc like, with enrichment in the large ion lithophile elements (LILEs), the light rare earth elements (LREEs), and P, along with depletion in the heavy rare earth elements (HREEs) and high‐field‐strength elements (HFSEs; e.g., Nb, Ta, Zr, Hf, and Ti). Whole‐rock (87Sr/86Sr)i ratios and εNd(t) values vary from 0.7037 to 0.7039 and from +7.62 to +8.45, respectively. The former is near equal to the 87Sr/86Sr ratios of plagioclase phenocrysts (0.7034–07039). Based on these geochemical data and the results of previous studies, we propose that the Sabeto shoshonitic primary magma was derived from partial melting of the depleted lithospheric mantle, modified by subduction‐related fluids. The thermobarometric calculation indicates that the P–T conditions of Fiji shoshonitic magma chamber are ~1100 °C and ~2.8 GPa (~90 km depth), with an average H2O fraction of ~6 wt%. In addition, the shoshonitic melts experienced fractional crystallization of clinopyroxene, olivine, and plagioclase during magma evolution. These volcanic rocks formed in an extensional setting following collision of the Vitiaz arc and the Ontong Java Plateau during the Pliocene. [ABSTRACT FROM AUTHOR]

A relaxation model employing double adiabatic magnetohydrodynamics (MHD) is introduced. In this model the neglect of heat flux in the Chew-Goldberger-Low (CGL) equations is treated by considering a very special class of the entropy invariants. The equilibria have uniform temperatures T[sub ⊥] and T[sub ], nearly uniform A ≡ j[sub ]/B = μ/σ, where σ is the local firehose parameter, and p = p (B). The density profile is not uniform (unless T[sub ⊥] = T[sub ]), in qualitative agreement with experiments. The equilibria are shown to be linearly stable, in spite of "bad" curvature, if the firehose and mirror parameters are positive and if the parallel current is not too large. A nonlinear stability criterion involving modified firehose and mirror parameters, and based upon the so-called energy-Casimir method, is also derived. [ABSTRACT FROM AUTHOR]

The self-focusing of relativistically intense laser light pulses is analyzed, where the pulse length is short enough that ion inertia prevents any significant motion of ions. Self-focusing occurs as a result of an increase of the wave refractive index arising from two effects: the mass increase of electrons caused by their relativistic quiver velocity in the light wave, and the reduction of the electron density as a result of ponderomotive force expulsion of the electrons. The latter effect is significant even for rather small values of (P-PL)/PL, where P is the laser beam power and PL is the critical power above which self-focusing occurs. In fact, for (P-PL)/PL>=0.1 the effect is so strong that all electrons are expelled within a core radial region of the self-focused laser light channel (this new phenomenon is called electron cavitation). [ABSTRACT FROM AUTHOR]

Previous work on the cubic direct interaction approximation applied to the truncated (in Fourier space) cubically nonlinear Schrödinger equation model of Langmuir turbulence is extended to more modes. In the undriven, undamped case, excellent agreement between the statistical theory and a numerical ensemble of solutions of the dynamic equations is obtained. In the driven, damped case, satisfactory agreement is obtained provided the dynamic ensemble is limited to initial conditions in the basin of attraction. [ABSTRACT FROM AUTHOR]

The cubic direct interaction approximation is applied to a truncated (in Fourier space) version of the cubically nonlinear Schrödinger equation model of Langmuir physics. The results are compared (in the three-mode case) to those for an ensemble of numerical solutions of the dynamical equations with 10000 different sets of Gaussianly distributed initial conditions. In the undriven, undamped case, the statistical theory (but not the ensemble) evolves to a state of thermal equilibrium. In the driven, damped case, the statistical theory appears to evolve to a state close to that corresponding to one of the limit cycles of the dynamical equations. [ABSTRACT FROM AUTHOR]

This is the first study of vocal individuality in male songs of black crested gibbons. The sound recordings were carried out at two field sites, Pinghe, Ailao Mountains, and Dazhaizi, Wuliang Mountains, both located in Yunnan province, China. A total of 127 coda phrases of 38 male songs bouts of eight individual male gibbons were analyzed. Stepwise discriminant function analysis was used to examine the acoustic individuality of the males. We found that individuality among neighbors was very pronounced. Moreover, individuality within a site (i.e. among neighbors) is higher than among individuals between sites. Our finding suggests that black crested gibbons may actively increase their degree of vocal individuality against that of their immediate neighbors by vocal adjustment., (© 2010 Wiley-Liss, Inc.)