IAS Benjamin Meaker Visiting Professor John Montgomery, University of Auckland, New Zealand

John Montgomery BMVP
Cerebellum-inspired flight control design for unmanned air vehicles

26 June - 14 July 2017

Professor Montgomery holds both a University of Bristol PhD and a DSc.  His PhD work was done at the Marine Biological Laboratory in Plymouth, supervised by Professors Denton FRS and Caldwell FRS.  His academic career has been at the University of Auckland where, until recently, he was Director of the Leigh Marine Laboratory, and the newly established Institute for Marine Science at the University. He is also a Principal Investigator in the Centre for Brain Research at the University of Auckland.  His scientific work sits at the interface of marine science and neuroscience and he has published extensively on sensory behaviour and physiology of fish, including hearing, hydrodynamic senses, and the quite extraordinary electrosensory system of sharks and rays.  The neuroscience context of his work includes the consideration of central mechanisms to distinguish signal and noise in sensory input, and the evolution of the cerebellum.  He has recently published a book with Oxford University Press “Evolution of the Cerebellar Sense of Self”.  This book explores the evolutionary origins of the cerebellum, and the generality of the ‘adaptive filter’ computational capabilities of the cerebellar circuitry.  His work is strongly interdisciplinary and has extensive overlap with engineering approaches and the use of biomimetics in engineering design.

His work has been recognised by election to the Royal Society of New Zealand, an International Brain Research Organisation Fellowship, Fulbright Scholarship, and James Cook Fellowship from the Royal Society of NZ.

Despite its relatively small size (10% by volume), the cerebellum has over 80% of the neurons in the human brain.  An evolutionary and neuroethological perspective of the cerebellum strongly implicates the cerebellum in the athleticism we see in swimming, flying and running vertebrates.  Where the cerebellar contribution includes: stabilization of the body core, and key sensors during locomotion; the anticipation of the sensory consequences of movement; and error driven correction of movement control.   All these contributions result from the adaptive filter computational capabilities of the cerebellar neural circuits. 

One way of viewing the cerebellar contribution is the generation of forward models from available control and sensory information to produce stabilization control, or predict expected sensory consequences.  As unmanned air vehicles (UAVs) becomes more sophisticated it is reasonable to propose that a cerebellum-like component will become a useful addition to flight control architecture. 

The next generation of UAVs are likely to carry many more sensors and potentially have morphing wings, where each part of the wing can be tailored to the local flow conditions to increase the endurance, robustness and utility of these vehicles.  However, these types of configurations are challenging to control with conventional control techniques, due to the large number of parallel input sensor streams and the flexibility of the control outputs required.  It is exactly this type of configuration that the cerebellum has evolved to control. In this project we will explore the potential use of cerebellum-like forward models in flight control for UAVs.   

During his stay in Bristol, Professor Montgomery will be hosted by Dr Shane Windsor (Engineering) and offer the following research events (Dates and times tbc):

Public Lecture: Cerebellar evolution and function: getting to know your cerebellar-self - Tuesday 27 June, 17:15, Small Lecture Theatre, Queen's Building, University Walk, Bristol, BS8 1TR

The cerebellum is an intriguing part of our brain. Its name is the diminutive form of cerebrum, so literally means ‘little brain’. It is true that, in humans, it occupies just 10% of the brain volume, yet recent research shows it accounts for approximately 80% of the nerve cells; a complex network of approximately 69 billion neurons! Why does the ‘little brain’ contain such a disproportionate number of neurons? How do we begin to understand the operation and function of such a complex network? What does the cerebellum do? And given that the cerebellum has the clear majority of our neurons, why is it so completely overshadowed by the cerebral cortex in psychology texts? An understanding of the cerebellum and what we might call the cerebellar ‘sense of self’ addresses these, and other, cerebellar questions. 

Departmental lecture
Basis function extension: the essence of a biological adaptive filter - Wednesday 5 July, 14:00-15:00, Queen's Building 1.18 LT
Dynamics and Control Research Group
In the adaptive filter model of the cerebellum, the granular layer performs a recoding which expands incoming mossy fibre signals into a temporally diverse set of basis signals. The underlying neural mechanism is not well understood, although various mechanisms have been proposed, including delay lines and echo state networks.  A computational simulation based on a network of leaky integrator neurons, and an adaptive filter performance measure, allows candidate mechanisms to be compared. Increasing circuit complexity improves adaptive filter performance, and this may relate to evolutionary innovations in the cerebellum and cerebellum-like structures in sharks and electric fish. Recurrence enables an increase in basis signal duration, which suggests a possible explanation for the explosion in granule cell numbers in the mammalian cerebellum. 

Graduate student seminar - Bristol Robotics Lab - Wednesday 28 June, 14:00-15:00, BRL Seminar Room (also broadcast to 2.59 Merchant Venturers Building)
Cerebellum: functional architecture from an evolutionary perspective
Abstract: The cerebellum occupies 10% of brain volume but contains nearly 80% of our neurons, and surprisingly we can literally live without it.  What is the functional architecture of this 68 billion neuronal network and what is its adaptive value?  The cerebellar circuitry first evolved in early vertebrates for discrimination of ‘self’ and ‘other’ in sensory systems.  The cancellation of self-generated sensory ‘noise’ is accomplished as a 2 step process: firstly by common-mode subtraction and then by a cerebellum-like adaptive filter circuitry.  This network uses copies of motor-command signals as a predictive set of basis functions.  The adaptive filter read-out element then uses a de-correlation learning rule to generate a forward model that is able to subtract the unwanted sensory consequences of self-generated movement.  The cerebellum itself appears to have arisen from this sensory adaptive filter.  It formed a ‘subsumption’ layer over the top of pre-existing motor control circuits, with adaptive filter functionality enhanced through various evolutionary innovations including: basis-function expansion; supervised learning; and clustering of output elements.  So the cerebellum can be thought of as a massive array of adaptive filters with multiple and diverse utility, and a contribution to behaviour that depends on the way in which it is wired into other operational circuits.