Dr Adam M Sykulski
|Themes||Multivariate and High Dimensional Data, Stochastic Modelling and Time Series|
As of 1st April 2017 I have moved to Lancaster University to become a Lecturer in Data Science and Statistics. I will however continue to interact with the UCL statistical science department as a visiting lecturer.
I am a statistician specialising in time series analysis, stochastic processes, and spatiotemporal data. I research practical problems that can not be analysed using the methods from a typical "Time Series 101" class. That means the data could be nonstationary, anisotropic, fractal or non-Markovian, and multivariate or high-dimensional.
I have an application focus in modelling large-scale global oceanographic data. I also investigate time series obtained from neuroscience and seismology, amongst others.
Please do get in touch if you have overlapping research interests.
I am an EU-funded Marie Curie Research Fellow. I spent the last two years working at NorthWest Research Associates in Seattle, USA - but I have now returned to the UCL Department of Statistical Science, as of April 2016.
My current research focuses primarily on developing stochastic process models and corresponding parameter estimation techniques for large-scale multivariate time series and spatiotemporal data. I have a particular interest in modelling in the frequency domain, and with using extensions of the Matern process. I have also developed an improved version of the Whittle likelihood, a quasi maximum likelihood procedure for the efficient parameter estimation of stochastic processes (see recent papers below).
I have applied my research to oceanographic data. I am implementing our methods on data from the Global Drifter Program: a large global database of satellite-tracked freely-drifting instruments known as 'drifters'. Our techniques allow us to make insightful new findings which improves global climate modelling and our ability to respond to environmental threats such as oil spills.
I also have an active research interest (from my PhD) in decision theory problems related to the multi-armed bandit problem, which is the simplest abstraction of the exploration-exploitation tradeoff. I have developed algorithms for how this tradeoff can be tuned on-line in practical problems. I have extended these ideas to multi-player problems, which then brings in ideas from game theory.
Here are slides available to download for a short course I give on the visualisation of spectral analysis methods. This is aimed at a senior undergraduate or graduate level:
Time Series Analysis, Stochastic Processes, Nonstationarity and Anisotropy, Applications in Oceanography, Decision Theory, Game Theory, Tennis.
Recent publications and preprints
- The de-biased Whittle likelihood for second-order stationary stochastic processes. Sykulski AM, Olhede SC and Lilly JM (2016). Link to ArXiv version
- Fractional Brownian motion, the Matérn process, and stochastic modeling of turbulent dispersion. Lilly JM, Sykulski AM, Early JJ and Olhede SC (2017). Nonlinear Processes in Geophysics (accepted as a discussion paper). Link to paper Link to ArXiv version
- Frequency-domain stochastic modeling of stationary bivariate or complex-valued signals. Sykulski AM, Olhede SC, Lilly JM and Early JJ (2017). IEEE Transactions on Signal Processing (accepted). Link to paper Link to ArXiv version
- Analysis of nonstationary modulated time series with applications to oceanographic flow measurements. Guillaumin AP, Sykulski AM, Olhede SC, Early JJ and Lilly JM (2017). Journal of Time Series Analysis, special issue on climate change (accepted). Link to ArXiv version
- Contribution to the discussion on "Should we sample a time series more frequently?: decision support via multirate spectrum estimation" by Nason et al. Sykulski AM (2017). Journal of the Royal Statistical Society Series A, 180(2), pp. 353-407. Link to paper
- Lagrangian time series models for ocean surface drifter trajectories. Sykulski AM, Olhede SC, Lilly JM and Danioux E (2016). Journal of the Royal Statistical Society Series C, 65(1), pp. 29-50. Link to paper Link to ArXiv version
- A widely linear improper complex autoregressive process of order one. Sykulski AM, Olhede SC and Lilly JM (2016). IEEE Transactions on Signal Processing, 64(23), pp. 6200-6210. Link to paper Link to ArXiv version
- A global surface drifter dataset at hourly resolution. Elipot S, Lumpkin R, Perez RC, Lilly JM, Early JJ and Sykulski AM (2016). Journal of Geophysical Research – Oceans, 121(5), pp. 2937-2966. Link to paper
- Exact simulation of noncircular or improper complex-valued stationary Gaussian processes using circulant embedding. Sykulski AM and Percival DB (2016). Proceedings of the 26th IEEE International Workshop on Machine Learning for Signal Processing, pp. 1 - 6, DOI: 10.1109/MLSP.2016.7738840. Link to paper Link to ArXiv version