Dynamic learning with the EM algorithm for neural networks

In this paper, we derive an EM algorithm for nonlinear state space models. We use it to estimate jointly the neural network weights, the model uncertainty and the noise in the data. In the E-step we apply a forwardbackward Rauch-Tung-Striebel smoother to compute the network weights. For the M-step, we derive expressions to compute the model uncertainty and the measurement noise. We find that the method is intrinsically very powerful, simple and stable.

Ferroelectric liquid crystal mixtures containing chiral ether and ester compounds with the 2-arylidene-p-menthan-3-one skeleton

Some 1R,4R-2-(4-phenylbenzylidene)-p-menthane-3-one derivatives containing the ether or ester linking group between benzene rings of the arylidene fragment have been studied as chiral dopants in ferroelectric liquid crystal systems based on the eutectic mixture (1:1) of two phenylbenzoate derivatives CmH2m+1OC6H4COOC6 H4OCnH2n+1 (n = 6; m = 8, 10). The ferroelectric properties of these compositions (spontaneous polarization, rotation viscosity, smectic tilt angle as well as quantitative characteristics of their concentration dependences) were compared with those for systems including chiral dopants containing no linking group. Ferroelectric parameters of the induced ferroelectric compositions studied have been shown to depend essentially on the presence of the linking group between benzene rings and its nature as well as on the number of the benzene rings in the rigid molecular core of the chiral dopants used. For all ferroelectric liquid crystal systems studied, the influence of the chiral dopants on the thermal stability of N*, SmA and SmC* mesophases has been quantified. The influence of the linking group nature in the dopant molecules on the characteristics of the systems studied is discussed taking into account results of the conformational analysis carried out by the semi-empirical AM1 and PM3 methods.

Entrepreneurial innovation problems associated with the dynamic growth of university spin-outs in China: A capabilities perspective

This research aims to develop a capabilities-based conceptual framework in order to study the stage-specific innovation problems associated with the dynamic growth process of university spin-outs (hereafter referred to as USOs) in China. Based on the existing literature, pilot cases and five critical cases, this study attempts to explore the interconnections between the entrepreneurial innovation problems and the configuration of innovative capabilities (that acquire, mobilise and re-configure the key resources) throughout the lifecycle of a firm in four growth phases. This paper aims to contribute to the literature in a holistic manner by providing a theoretical discussion of USOs' development through adding evidence from a rapid growth emerging economy. To date, studies that have investigated the development of USOs in China recognised the heterogeneity of USOs in terms of capabilities still remain sparse. Addressing this research gap will be of great interest to entrepreneurs, policy makers and venture investors. © Copyright 2010 Inderscience Enterprises Ltd.

Mode division multiplexing of modes with the same azimuthal index

The design of an SLM-based mode demultiplexer is discussed and mode division multiplexing is performed using the LP0,1 and LP 0,2 modes, representing the first demonstration to propagate channels on modes with the same azimuthal index. Mode multiplexed transmission over 2 km of 50-μm OM2 fiber demonstrates a modal selectivity of 16 dB and an OSNR penalty of 1.5 dB for the transmission of 2×56 Gb/s QPSK signals. © 2012 IEEE.