



I am interested in the analysis and design of algorithms that extract actionable information from sensor data. Examples include surveillance systems in which case these algorithms are often said to provide "situation awareness", and, autonomous systems in which case this capability is almost synonymous with machine perception. Typically these systems collect measurements from the environment using sensors such as radars, cameras, acoustic and environmental sensors. Inference based upon sensor data involves modelling the evolution of the "world", the sensing system as well as the sensors and updating these models as new measurements arrive using frameworks from, for example, dynamical systems theory, the theory of probability, optimisation and computational statistics.

In this scenario, there are two important processing challenges that motivate my research: The first is the limited resources (e.g. computational budget, access to storage, communication bandwidth etc.) available to update a sufficiently complex environment model at the rate of such data streams (e.g., tens of millions of complex numbers per second in the case of a staring array radar  a similar rate in the case of laser range finders in autonomous driving applications). The second challenge is related to that complex environments by definition cannot be modelled by building upon naive assumptions. These models ought to be learned from and/or verified using previously recorded data. Such training data, however, is a scarce and expensive resource. Data scarce learning in sensing of dynamical systems hence merits research for futureproof sensing capability.

My research objective is to contribute to the understanding and methodological aspects of these challenges for general cybernetic systems.

The fields of study I am interested in are:
Statistical inference, signal processing and information fusion; Bayesian paradigms; Variational approaches, probabilistic
graphical models and message passing algorithms; Dynamic system models, population processes (e.g. Random finite sets);
Machine learning (from a statistical perspective)

Coupled with these fields are the following themes:
Multiscale inference in dynamical models (i.e. in multiple levels of abstraction); Highlevel information driven lowlevel sensor signal processing and fusion (read, inference in state space models with tensor valued measurements), massively parallelisable inference in multisensor multiobject settings, resource use (e.g. communication bandwidth) and accuracy tradeoffs in detection and estimation

Please check
my publications each of which is built upon a combination of the above elements.




