Parameter estimation for agenda-based user simulation

This paper presents an agenda-based user simulator which has been extended to be trainable on real data with the aim of more closely modelling the complex rational behaviour exhibited by real users. The train-able part is formed by a set of random decision points that may be encountered during the process of receiving a system act and responding with a user act. A sample-based method is presented for using real user data to estimate the parameters that control these decisions. Evaluation results are given both in terms of statistics of generated user behaviour and the quality of policies trained with different simulators. Compared to a handcrafted simulator, the trained system provides a much better fit to corpus data and evaluations suggest that this better fit should result in improved dialogue performance. © 2010 Association for Computational Linguistics.

Parameter optimization for automated concrete detection in image data

Several research studies have been recently initiated to investigate the use of construction site images for automated infrastructure inspection, progress monitoring, etc. In these studies, it is always necessary to extract material regions (concrete or steel) from the images. Existing methods made use of material's special color/texture ranges for material information retrieval, but they do not sufficiently discuss how to find these appropriate color/texture ranges. As a result, users have to define appropriate ones by themselves, which is difficult for those who do not have enough image processing background. This paper presents a novel method of identifying concrete material regions using machine learning techniques. Under the method, each construction site image is first divided into regions through image segmentation. Then, the visual features of each region are calculated and classified with a pre-trained classifier. The output value determines whether the region is composed of concrete or not. The method was implemented using C++ and tested over hundreds of construction site images. The results were compared with the manual classification ones to indicate the method's validity.

Physical consequences of a nonparametric uncertainty model in structural dynamics

One of the main claims of the nonparametric model of random uncertainty introduced by Soize (2000) [3] is its ability to account for model uncertainty. The present paper investigates this claim by examining the statistics of natural frequencies, total energy and underlying dispersion equation yielded by the nonparametric approach for two simple systems: a thin plate in bending and a one-dimensional finite periodic massspring chain. Results for the plate show that the average modal density and the underlying dispersion equation of the structure are gradually and systematically altered with increasing uncertainty. The findings for the massspring chain corroborate the findings for the plate and show that the remote coupling of nonadjacent degrees of freedom induced by the approach suppresses the phenomenon of mode localization. This remote coupling also leads to an instantaneous response of all points in the chain when one mass is excited. In the light of these results, it is argued that the nonparametric approach can deal with a certain type of model uncertainty, in this case the presence of unknown terms of higher or lower order in the governing differential equation, but that certain expectations about the system such as the average modal density may conflict with these results. © 2012 Elsevier Ltd.

Uncertainty management for on-line optimisation of a POMDP-based large-scale spoken dialogue system

The optimization of dialogue policies using reinforcement learning (RL) is now an accepted part of the state of the art in spoken dialogue systems (SDS). Yet, it is still the case that the commonly used training algorithms for SDS require a large number of dialogues and hence most systems still rely on artificial data generated by a user simulator. Optimization is therefore performed off-line before releasing the system to real users. Gaussian Processes (GP) for RL have recently been applied to dialogue systems. One advantage of GP is that they compute an explicit measure of uncertainty in the value function estimates computed during learning. In this paper, a class of novel learning strategies is described which use uncertainty to control exploration on-line. Comparisons between several exploration schemes show that significant improvements to learning speed can be obtained and that rapid and safe online optimisation is possible, even on a complex task. Copyright © 2011 ISCA.

The parameter space of graphene chemical vapor deposition on polycrystalline Cu

A systematic study of the parameter space of graphene chemical vapor deposition (CVD) on polycrystalline Cu foils is presented, aiming at a more fundamental process rationale in particular regarding the choice of carbon precursor and mitigation of Cu sublimation. CH 4 as precursor requires H 2 dilution and temperatures ≥1000 °C to keep the Cu surface reduced and yield a high-quality, complete monolayer graphene coverage. The H 2 atmosphere etches as-grown graphene; hence, maintaining a balanced CH 4/H 2 ratio is critical. Such balance is more easily achieved at low-pressure conditions, at which however Cu sublimation reaches deleterious levels. In contrast, C 6H 6 as precursor requires no reactive diluent and consistently gives similar graphene quality at 100-150 °C lower temperatures. The lower process temperature and more robust processing conditions allow the problem of Cu sublimation to be effectively addressed. Graphene formation is not inherently self-limited to a monolayer for any of the precursors. Rather, the higher the supplied carbon chemical potential, the higher the likelihood of film inhomogeneity and primary and secondary multilayer graphene nucleation. For the latter, domain boundaries of the inherently polycrystalline CVD graphene offer pathways for a continued carbon supply to the catalyst. Graphene formation is significantly affected by the Cu crystallography; i.e., the evolution of microstructure and texture of the catalyst template form an integral part of the CVD process. © 2012 American Chemical Society.

Probabilistic computation in human perception under variability in encoding precision.

A key function of the brain is to interpret noisy sensory information. To do so optimally, observers must, in many tasks, take into account knowledge of the precision with which stimuli are encoded. In an orientation change detection task, we find that encoding precision does not only depend on an experimentally controlled reliability parameter (shape), but also exhibits additional variability. In spite of variability in precision, human subjects seem to take into account precision near-optimally on a trial-to-trial and item-to-item basis. Our results offer a new conceptualization of the encoding of sensory information and highlight the brain's remarkable ability to incorporate knowledge of uncertainty during complex perceptual decision-making.

Two problems with variational expectation maximisation for time-series models

Variational methods are a key component of the approximate inference and learning toolbox. These methods fill an important middle ground, retaining distributional information about uncertainty in latent variables, unlike maximum a posteriori methods (MAP), and yet generally requiring less computational time than Monte Carlo Markov Chain methods. In particular the variational Expectation Maximisation (vEM) and variational Bayes algorithms, both involving variational optimisation of a free-energy, are widely used in time-series modelling. Here, we investigate the success of vEM in simple probabilistic time-series models. First we consider the inference step of vEM, and show that a consequence of the well-known compactness property of variational inference is a failure to propagate uncertainty in time, thus limiting the usefulness of the retained distributional information. In particular, the uncertainty may appear to be smallest precisely when the approximation is poorest. Second, we consider parameter learning and analytically reveal systematic biases in the parameters found by vEM. Surprisingly, simpler variational approximations (such a mean-field) can lead to less bias than more complicated structured approximations.