Sequential Monte Carlo is an emerging field. I assume that the paper is highly cited because a large number of scientists are interested in learning and applying Sequential Monte Carlo to their problems.

  Does it describe a new discovery or new methodology that's useful to others?

It includes in a general and simple framework many algorithms proposed independently in the literature. Based on this framework, new algorithms are also proposed.

This research deals with optimal filtering; i.e., how to estimate optimally a stochastic process given some noisy observations. It has numerous applications in control, signal processing, finance, computer vision, genetics, robotics, atmospheric sciences etc.

For most realistic stochastic models, this problem does not admit a closed-form solution. For over 30 years, many approximation schemes, such as the extended Kalman filter, Gaussian sum approximations and grid-based filters have been proposed to surmount this problem. The first two methods fail to take into account all the salient statistical features of the models whereas grid-based algorithms, based on deterministic numerical integration methods, can lead to accurate results, but are difficult to implement and too computationally intensive to be of any practical interest in high dimensions. Sequential Monte Carlo have proved to be an elegant, efficient, flexible and powerful way to perform these approximations.

Many problems in applied science can be casted as an optimal filtering problem; i.e., estimating the volatility of the stock market, demodulating a communication signal, weather forecasting etc. In most real-world models, the optimal filter does not admit a closed-form solution. Sequential Monte Carlo are a set of powerful numerical methods which can arbitrarily closely approximate the solution. They do not rely on any simplifying assumption, are extremely flexible and they can be easily implemented on parallel computers. Many nonlinear non-Gaussian estimation problems which were out of reach only 5 years ago can now be solved efficiently.

Dr. Arnaud Doucet
Department of Electrical Engineering, Melbourne University,
Parkville, 3010 Victoria, Australia.