Bayesian Model

The innovative focus of the BIBA project is the coordinated application of probabilistic
(Bayesian) inference and learning to the problems of behavioural control in animals and
artefacts. The real world - especially in the aspects that animals excel at handling - is
characterised by uncertainty, unpredictability and noise. There are few if any problems of
control and behaviour that should not be looked at with this in mind, yet the current state of
the art does not give this the overriding importance it deserves. Bayesian inference is the
necessary extension of logic in these circumstances, when there is insufficient information to
perceive, infer, decide and act on rigorous principles. Though many other groups do include
probabilistic reasoning in their thinking, we see the BIBA project - with committed experts
from different backgrounds, focussed on this issue - as an opportunity to make a synthesis of
ideas in biology and engineering, with each inspiring the other by exposure to novel aspects
of the problems and novel approaches, thus helping to consolidate a genuine paradigm shift.

Ever since the initial work of Laplace and Bayes, probability theory has been regarded as a
natural way to deal with incomplete models of natural phenomenon. Recently, the Bayesian
approach has been taken forward strongly by the late Edward T. Jaynes, with a rigorous
synthetic formulation of probabilistic reasoning, sometimes encapsulated by the phrase
"Probability as Logic". Any model of a complex phenomenon is bound to be irremediably
incomplete. There are always some "hidden" variables not taken into account. Consequently,
the phenomenon and the model never behave exactly alike. The purpose of Bayesian
inference and learning is to optimise reasoning in such a context. The BIBA project will
transform this formal approach to a set of tools and practices, which could be shared by
scientists coming from two communities: life science and engineering.