In this paper we present a graphical model for polyphonic music transcription. Our model, formulated as a Dynamical Bayesian Network, embodies a transparent and computationally tractable approach to this acoustic analysis problem. An advantage of our approach is that it places emphasis on explicitly modelling the sound generation procedure. It provides a clear framework in which both high level (cognitive) prior information on music structure can be coupled with low level (acoustic physical) information in a principled manner to perform the analysis. The model is a special case of the, generally intractable, switching Kalman filter model. Where possible, we derive, exact polynomial time inference procedures, and otherwise efficient approximations. We argue that our generative model based approach is computationally feasible for many music applications and is readily extensible to more general auditory scene analysis scenarios.

This paper looks at unobserved components models and examines the implied weighting patterns for signal extraction. There are three main themes. The first is the implications of correlated disturbances driving the components, especially those cases in which the correlation is perfect. The second is how setting up models with t \Gamma distributed disturbances leads to weighting patterns which are robust to outliers and breaks. The third is a comparison of implied weighting patterns with kernels used in nonparametric trend estimation and equivalent kernels used in spline smoothing. We also examine how weighting patterns are affected by heteroscedasticity and irregular spacing and provide an illustrative example.

The actual effort to evolve and maintain a software system is likely to vary depending on the amount of clones (i.e., duplicated or slightly different code fragments) present in the system.

Many different short-time features, using time windows in the size of 10-30 ms, have been proposed for music segmentation, retrieval and genre classification. However, often the available time frame of the music to make the actual decision or comparison (the decision time horizon) is in the range of seconds instead of milliseconds. The problem of making new features on the larger time scale from the short-time features (feature integration) has only received little attention. This paper investigates different methods for feature integration and late information fusion 1 for music genre classification. A new feature integration technique, the AR model, is proposed and seemingly outperforms the commonly used meanvariance features. 1.

We take a nonstructural time-series approach to modeling and forecasting daily average temperature in ten U.S. cities, and we inquire systematically as to whether it may prove useful from the vantage point of participants in the weather derivatives market. The answer is, perhaps surprisingly, yes. Time series modeling reveals both strong conditional mean dynamics and conditional variance dynamics in daily average temperature, and it reveals sharp differences between the distribution of temperature and the distribution of temperature surprises. Most importantly, it adapts readily to produce the long-horizon forecasts of relevance in weather derivatives contexts. We produce and evaluate both point and distributional forecasts of average temperature, with some success. We conclude that additional inquiry into nonstructural weather forecasting methods, as relevant for weather derivatives, will likely prove useful. Key Words: Risk management; hedging; insurance; seasonality; average temperature; financial derivatives; density forecasting JEL Codes: G0, C1 Acknowledgments: For financial support we thank the National Science Foundation, the Wharton Financial Institutions Center, and the Wharton Risk Management and Decision Process Center. For helpful comments we thank Marshall Blume, Larry Brown, Jeff Considine, John Dutton, Ren Garcia, Stephen Jewson, Vince Kaminski, Paul Kleindorfer, Howard Kunreuther, Yu Li, Bob Livezey, Cliff Mass, Don McIsaac, Nour Meddahi, David Pozo, Matt Pritsker, S.T. Rao, Claudio Riberio, Til Schuermann and Yihong Xia. We are also grateful for comments by participants at the American Meteorological Society's Policy Forum on Weather, Climate and Energy. None of those thanked, of course, are responsible in any way for the outcome. Address corresponde...

This paper reports empirical evidence that a neural network model is applicable to the prediction of foreign exchange rates. Time series data and technical indicators, such as moving average, are fed to neural networks to capture the underlying rulesa of the movement in currency exchange rates. The exchange rates between American Dollar and "ve other major currencies, Japanese Yen, Deutsch Mark, British Pound, Swiss Franc and Australian Dollar are forecast by the trained neural networks. The traditional rescaled range analysis is used to test the e$ciencya of each market before using historical data to train the neural networks. The results presented here show that without the use of extensive market data or knowledge, useful prediction can be made and signi"cant paper pro"ts can be achieved for out-of-sample data with simple technical indicators. A further research on exchange rates between Swiss Franc and American Dollar is also conducted. However, the experiments show that with e$cient market it is not easy to make pro"ts using technical indicators or time series input neural networks. This article also discusses several issues on the frequency of sampling, choice of network architecture, forecasting periods, and measures for evaluating the model's predictive power. After presenting the experimental results, a discussion on future research concludes the paper. # 2000 Elsevier Science B.V. All rights reserved.

For many dynamic estimation problems involving nonlinear and/or non-Gaussian models, particle filtering offers improved performance at the expense of computational effort. This paper describes a scheme for efficiently tracking multiple targets using particle filters. The tracking of the individual targets is made efficient through the use of Rao-Blackwellisation. The tracking of multiple targets is made practicable using Quasi-Monte Carlo integration. The efficiency of the approach is illustrated on synthetic data.

We describe methods for applying Monte Carlo filtering and smoothing for estimation of unobserved states in a nonlinear state-space model. By exploiting the statistical structure of the model, we develop a Rao--Blackwellized particle smoother. Due to the lengthy nature of real signals, we suggest processing the data in blocks, and a block-based smoother algorithm is developed for this purpose. All the algorithms suggested are tested with real speech and audio data, and the results are shown and compared with those generated using the generic particle smoother and the extended Kalman filter (EKF). It is found that the proposed Rao--Blackwellized particle smoother improves on the standard particle smoother and the extended Kalman smoother. In addition, the proposed Block-based smoother algorithm enhances the efficiency of the proposed Rao--Blackwellized smoother by significantly reducing the storage capacity required for the particle information.

This paper reviews the general Bayesian approach to parameter estimation in stochastic volatility models with posterior computations performed by Gibbs sampling. The main purpose is to illustrate the ease with which the Bayesian stochastic volatility model can now be studied routinely via BUGS (Bayesian Inference Using Gibbs Sampling), a recently developed, user-friendly, and freely available software package. It is an ideal software tool for the exploratory phase of model building as any modifications of a model including changes of priors and sampling error distributions are readily realized with only minor changes of the code. However, due to the single move Gibbs sampler, convergence can be slow. BUGS automates the calculation of the full conditional posterior distributions using a model representation by directed acyclic graphs. It contains an expert system for choosing an effective sampling method for each full conditional. Furthermore, software for convergence diagnostics and statistical summaries is available for the BUGS output

Dans ce papier, on jauge l'utilit de plusieurs estimations (univaries autant que multivaries) de l'cart de production pour prvoir le taux d'inflation. Une analyse ex post suggre que plusieurs de ces estimations aident prdire l'inflation. Nanmoins, les erreurs de prdictions hors de l'enchantillon qui se sont construites avec les carts de production estims en temps rel indiquent que cette amlioration de prdiction est illusoire. On trouve que l'utilit des carts de production pour prdire l'inflation a t exagre et que les prdictions faites avec l'cart de production sont souvent moins prcises que celles qui ignorent le concept d'un cart de production.