Call classification using recurrent neural networks, support vector machines and finite state automata

Our objective is spoken-language classification for helpdesk call routing using a scanning understanding and intelligent-system techniques. In particular, we examine simple recurrent networks, support-vector machines and finite-state transducers for their potential in this spoken-language-classification task and we describe an approach to classification of recorded operator-assistance telephone utterances. The main contribution of the paper is a comparison of a variety of techniques in the domain of call routing. Support-vector machines and transducers are shown to have some potential for spoken-language classification, but the performance of the neural networks indicates that a simple recurrent network performs best for helpdesk call routing. Sheila Garfield received a BSc (Hons) in computing from the University of Sunderland in 2000 where, as part of her programme of study, she completed a project associated with aphasic language processing. She received her PhD from the same university, in 2004, for a programme of work connected with hybrid intelligent systems and spoken-language processing. In her PhD thesis, she collaborated with British Telecom and suggested a novel hybrid system for call routing. Her research interests are natural language processing, hybrid systems, intelligent systems. Stefan Wermter holds the Chair in Intelligent Systems and is leading the Intelligent Systems Division at the University of Sunderland, UK. His research interests are intelligent systems, neural networks, cognitive neuroscience, hybrid systems, language processing and learning robots. He has a diploma from the University of Dortmund, Germany, an MSc from the University of Massachusetts, USA, and a PhD in habilitation from the University of Hamburg, Germany, all in Computer Science. He was a Research Scientist at Berkeley, CA, before joining the University of Sunderland. Professor Wermter has written edited, or contributed to 8 books and published about 80 articles on this research area.     

Autonomic correlates of attention-deficit/hyperactivity disorder and oppositional defiant disorder in preschool children.

Numerous studies have revealed autonomic underarousal in conduct-disordered adolescents and antisocial adults. It is unknown, however, whether similar autonomic markers are present in at-risk preschoolers. In this study, the authors compared autonomic profiles of 4- to 6-year-old children with attention-deficit/hyperactivity disorder (ADHD) and oppositional defiant disorder (ODD; n = 18) with those of age-matched controls (n = 20). Children with ADHD and ODD exhibited fewer electrodermal responses and lengthened cardiac preejection periods at baseline and during reward. Although group differences were not found in baseline respiratory sinus arrhythmia, heart rate changes among ADHD and ODD participants were mediated exclusively by parasympathetic withdrawal, with no independent sympathetic contribution. Heart rate changes among controls were mediated by both autonomic branches. These results suggest that at-risk preschoolers are autonomically similar to older externalizing children. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Personal reflections: Barriers and strategies in increasing diversity in psychology.

In this article, six faculty and students of color who participated in a panel discussion at a symposium during the National Multicultural Conference and Summit of 2003 talk about the barriers they encountered and continue to encounter in their graduate training and places of employment. They also discuss strategies they found to be effective, enhancing, and positive and suggest other possibilities. The contributors describe their relationships with dominant-group and minority peers and talk about how issues of social class, disability, and sexual orientation as well as color have been part of their experience. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Statistical considerations in infant nutrition trials

Infant nutrition trials usually require developmental follow-up, often to 18 mon, and sometimes beyond reading ability at age 7 yr. They are therefore logistically complex and costly, and should be conducted to a high statistical standard. With examples, we focus on: good practice in nutrition trials and the goal of a common protocol; how to set plausible trial targets and to work out trial size accordingly; statistical observations on assessing visual maturation; and methods of randomization, including the method known as minimization, which can be adapted to select more appropriately a comparison cohort of breast-fed infants. We end with discussion of analysis and reporting standards, including the give-aways (or tell-tale signs) to be on the look out for. A Cochrane Collaboration for systematic review of randomized trials of infant nutrition is proposed.     

Demonstration and analysis of a 40-gigasample/s interferometric analog-to-digital converter

A novel interferometric scheme for photonic analog-to-digital conversion is for the first time experimentally demonstrated at a real-time sample rate of 40 gigasamples/s. The scheme includes sampling as well as binary encoding, and the input signal in the experiment was a 1.25-GHz sinusoidal tone that was successfully digitized with a nominal resolution of 21 digital levels. Single-sample measurements yielded an effective number of bits (ENOB) of 2.6, which was limited by thermal detection noise while multisample averaged measurements resulted in an ENOB of 3.6 bits, mainly limited by phase drift. Apart from the experimental results, this paper covers an extensive theoretical analysis of the system, including calculations on the fundamental maximum bandwidth, the required optical power, the generated binary code, and its error robustness, as well as the impact of detection noise on the signal-to-noise ratio of the digitized signal. The major benefits of this interferometric scheme are that only one standard phase modulator is required and that the phase swing does not have to be larger than /spl plusmn//spl pi/ to reach the full digital value space.     

Estimation of the exposure of the French population to the BSE agent: comparison of the 1980-95 consumption of beef products containing mechanically recovered meat in France and the UK,by birth cohort and gender

Assuming that human exposure to BSE was through beef mechanically recovered meat (MRM) consumed as burgers and other meat products, we estimated the French consumption of different food items containing beef MRM, and compared these consumptions for French and British populations. To estimate consumption of meat products containing bovine MRM, we used dietary data from national individual and household food surveys conducted between 1980 and 1995. After reconciliation of consumption data between the available surveys and calendar year adjustments, we simulated consumption of one-thousandth of the French population. Consumption was estimated by birth cohort and gender, and for the periods 1980-89 and 1990-95 separately. Data showed that burgers (including manufactured minced meat) represented around 75-80% of the individual consumption of meat products containing MRM, and that consumption of burgers increased by 40% over the 1980-95 period. In all age groups, consumption was higher in males than in females. In both genders, the 1940-69 birth cohort had the highest mean consumption of burgers and other beef products containing MRM. Similar findings have been reported for the UK population. Estimated consumption of bovine MRM per calendar year increased markedly over the study period, concomitantly with an increase of bovine carcasses imported from the UK. Comparison of the 1980-1995 pattern of bovine MRM consumption in the UK and France indicated thatthis consumption peaked later in France than in the UK. This difference might result in different temporal pattern of vCJD incidence.     

Dynamic demonstration of diffractive optic analog-to-digital converter scheme

Dynamic behavior of an analog-to-digital converter (ADC) based on diffractive optical element(s) (DOE)(s) was studied and found to be in agreement with predictions. The analog signal was translated to an angular deflection of a laser beam by means of an acousto-optic (AO) cell. The number of bits in this experimental demonstration was three, using an eight-element DOE array. The maximum sample rate was found to be 2.5 MS/s, the limiting factor being the transit time for the acoustic wave across the width of the laser beam in the AO cell. The study is intended as a first dynamic demonstration of a proposed ADC scheme previously demonstrated in a quasi-static version. The full potential of the ADC scheme will require the use of a fast tunable diode laser to replace the AO deflection scheme used here.     

Efficient entropy estimation based on doubly stochastic models for quantized wavelet image data.

Under a rate constraint, wavelet-based image coding involves strategic discarding of information such that the remaining data can be described with a given amount of rate. In a practical coding system, this task requires knowledge of the relationship between quantization step size and compressed rate for each group of wavelet coefficients, the R-Q curve. A common approach to this problem is to fit each subband with a scalar probability distribution and compute entropy estimates based on the model. This approach is not effective at rates below 1.0 bits-per-pixel because the distributions of quantized data do not reflect the dependencies in coefficient magnitudes. These dependencies can be addressed with doubly stochastic models, which have been previously proposed to characterize more localized behavior, though there are tradeoffs between storage, computation time, and accuracy. Using a doubly stochastic generalized Gaussian model, it is demonstrated that the relationship between step size and rate is accurately described by a low degree polynomial in the logarithm of the step size. Based on this observation, an entropy estimation scheme is presented which offers an excellent tradeoff between speed and accuracy; after a simple data-gathering step, estimates are computed instantaneously by evaluating a single polynomial for each group of wavelet coefficients quantized with the same step size. These estimates are on average within 3% of a desired target rate for several of state-of-the-art coders.