Company Background
Vector Anomaly was founded by Mark Hatton and Mike Tipping in 2006. Prior to that date, we had been busy establishing a genuine world-class
reputation for innovation in data-analytic science, both in academia and in industry. As a result, over the past ten years, we have
acquired an exceptional level of specialist knowledge spanning a
number of contemporary research fields (see our Technology section) which are highly relevant to applied statistical analysis and data modelling.
During this time we have been responsible for a number of novel and
significant contributions in data analysis and related fields
(a selection of relevant publications may be found on
our Science page). Methods such as "probabilistic
principal component analysis",
for example, have become well-known. Furthermore, we originated and
developed the concept of "sparse Bayesian" modelling and
invented the popular "relevance vector
machine". This novel
technology for the automated development of efficient Bayesian prediction
models continues to have widespread international impact, and is already being exploited in a diverse range
of applications, including glaucoma diagnosis, gene classification,
visual object tracking and financial market forecasting.
While we feel that we may be justifiably proud of our record in pure
research, we have also acquired a wealth of experience of practical
application development. We have had considerable success, both
directly and in a consultancy role, in developing effective solutions
for a variety of internal and external clients across a broad range of application areas. This includes bioinformatics, financial
database analysis, handwriting recognition, pharmaceutical
manufacturing, vehicle control, energy transmission systems modelling
and interactive entertainment. At the same time, this has also led to us directly
developing substantial components of commercial software with users
numbering in the millions.
Philosophy
Our overall philosophy towards data analysis, problem solving and system
implementation is best summarised as principled, practical and probabilistic.
Most importantly, we
recognise that in the real-world application environment, measured data is not
100% reliable, user-intepreted data even less so, systems are not completely
deterministic, software implies bugs, and specified queries are not necessarily
easily categorised into binary "yes" and "no" answers. Through diligent application of
a principled probabilistic methodology, our techniques treat error and uncertainty in a
scientific manner in order to maximise overall system fidelity and to avoid the
implication of a fallacious level of confidence. Overall, we endeavour to
design tractable and elegant solutions which are based on sound mathematical
principles and yet remain grounded in implementational practicality.
People
You can read our individual "biographies" by clicking on our pictures below.
|
