David Howard

"The beauty of a living thing is not the atoms that go into it, but the way they are put together."
Carl Sagan


I'm research associate at the University of the West of England. Before I came to Bristol, I spent around 4 years in Leeds where I earned a BSc in Computing and MSc in Cognitive Systems. Throughout my BSc and MSc I became especially inspired by the field of biologically-inspired computing - this can be best thought of as observing some facet of nature, forming a working abstraction of a natural process (for example, ant pheremone behaviour), adapting it for use of a computer, and using it to solve problems in ways that more traditional methods can't.

Inspiration for much of my work implicitly owes a lot to Darwinian theories of evolution. For example, selection pressure - the infamous "survival of the fittest" - is used to keep high-quality individuals in the population whilst killing off individuals that are of little utility. In addition, abstractions of both sexual and asexual reproduction can be seen as a means of breeding and enhancing desirable traits within certain rules.

My main area of application is in autonomous robotics - robots that can flexibly learn and adapt to their surroundings by continuously updating their knowledge base in a self-guided manner. Extensions include features such as self-modifying memory and self-repair in the event of hardware failure.


Running, cycling when I get the chance, football and american football. I'm a fan of York City and the New England Patriots.

I also read a lot; some science/biography, mainly sci-fi and classics. I'm also interested in ancient Greek and Roman culture.

I enjoy the chance to travel and present/discuss my work at various conferences as well as popular science events, having been an invited speaker at the Wrexham Science Festival (slides and videos available below).


The pervasive motif of my work to date has been in adaptive control, including autonomous learning and neuro-evolution. My controllers are tested in simulation and hardware on both stationary and dynamic environments.

Research Associate

I'm currently investigating memory-resistors, or memristors, and their application in various network paradigms. A memristor is a next-generation nanoscale memory candidate whose instantaneous resistance depends on the entire history of charge that has passed through it. As the charge affects the chemical properties of the device, the resistance is nonvolatile and will persist even when the power source is disconnected. The main benefit of evolving memristor ciruits is that, as the models used were based on physical devices, any evolved simulated designs could be recreated in hardware, with the potential for extension into evolvable hardware.

Evolution of a plethora of network types, for example Random Boolean Networks, A-type Unorganised Machines, and Cartesian Genetic Programming grids have been investigated. Physical memristors have been used as part of "in the loop" evolution of logical circuits. Detailed simulation models have also been created from measurements of physical devices - extensive modelling of physical memristors has led to the creation of a model of the PEO-PANI memristor for the SPICE simulation program.

The main result of this research so far is a result of the discovery that the properties of a memristor (nonvolatility, charge-dependent state) make it an ideal synapse analog in a neural network, whereby Hebbian learning can permanently alter the conductivity of the synapse, until the next STDP event takes place. Initial work has demonstrated a prototype Neuromorphic ("brain-like") system including neuro-evolution underpinned with self-adaptive learning control, which has been deployed for unsupervised robotic obstacle avoidance and memory tasks.


My PhD was focussed on population-based reinforcement learning systems called Learning Classifier Systems (LCS), using spiking neural networks in place of the traditional representation. These online systems employed constructive neuro-evolution to create full, accurate payoff maps via Q-learning. My thesis includes the first demonstration of continuous valued actions in continuous space and time on an LCS, as well as a method for automatically discretising state space into required basic behaviours by chaining together actions into single "macro-actions". This approach was found to significantly reduce the required number of classifiers for the robotics obstacle avoidance/phototaxis tasks considered.

Member of IEEE CEC review committee

Member of ACM GECCO GA and GBML programme committees

Member of UARACIN programme committee

Reviewer for IEEE Transactions on Evolutionary Computing, IEEE Transactions on Circuits and Systems

Presented at the Wrexham Science Festival 2010

PhD thesis

MSc thesis

Journal Articles

Conference Proceedings

Selected technical reports

Academic/Formal Contact

email: david4DOThowardATuweDOTacDOTuk

Informal Contact

email: gdhowardAThotmailDOTcoDOTuk