Bath AI Seminar Series: 2009

 


Bath Artificial Intelligence (BAI) Seminar is a research and discussion group. Announcements are made in the BAI mailing list, which also has its own archive.

Directions to Bath and a campus map: Computer Science is in 1W, beside the library.  Upcoming talks are listed below and announced by email.

Date Location Guest(s) Topic
22 Oct 3W 3.7 Benjamin Williamson & William Megill Seabiscuit, Bath's 2009 Autonomous Underwater Vehicle
29 Oct 3W 3.7 Rizos Sakellariou Decision making for autonomic solutions on the Grid
19 Nov 3W 3.7 David Gamez The Development and Analysis of Conscious Machines

The talk will start by situating work on machine consciousness within
the broader framework of the scientific study of consciousness. The
specific problems raised by machine consciousness will then be
covered, along with the extent to which we can make accurate
predictions about the consciousness of machines. Next, experimental
work will be summarized, in which a spiking neural network was used to
implement models of imagination and emotion that controlled the eye
movements of the SIMNOS virtual robot. This network had several
cognitive functions that have been linked to consciousness, and
predictions were made about the consciousness of the network using the
theories of Tononi, Aleksander and Metzinger. The talk will conclude
with some of the applications of this work to the scientific study of
consciousness and cognitive engineering.
20 November 1W 3.6 Joanna Bryson A Primer on AI for Domestic Robots: Does Thinking Help?
26 Nov 3W 3.7 Marios Richards Comparing individuality to phenotypic plasticity in facilitating evolution

Natural selection requires variation to operate – there is no change without options to select between.  Variation is therefore itself subject to selection, including its rate and extent.  Individual plasticity (including cognition) is known to facilitate evolution in two ways.  First, it accelerates evolution when a species is far enough from its local optima that even unreliable processes like individual learning provide evidence that genetic evolution can exploit about which animals are closer to optimal. But second, plasticity also slows genetic convergence once the adaptive phenotype is close enough to genetic specification that individuals can reliably learn to be `fit’.  This reliability reduces selective pressure, and thus retains diversity. Here we present an extension to a classic model of the interaction between plasticity & natural selection (Hinton & Nowlan 1987), to examine whether increased genetic variation in offspring can play a similar role to individual plasticity. We find that a higher recombination rate significantly accelerates the earlier phase of the model, where the organism is ill-suited to its environment and deleterious traits require elimination from the genome.  This indicates that genetic variation plays a role similar to plasticity in this stage.  We speculate that a behavioural search for identity in social species may play a similar role. 
10 Dec 3W 3.7 Etienne Roesch Attention, emotion and neural networks

In a conscious experience, attention selects the elements in the
environment that will undergo further processing. Emotion influences this
mechanism by evaluating the significance of the stimuli perceived, and by
contributing to the ensuing phenomenology. I investigated the intricacies of
this system through two experimental paradigms: the modulation of the
attentional blink by emotion-laden information, and a psychophysical
paradigm studying the mechanisms responsible for the automatic processing of
this information. I used the results of this work to inform theories, and
sketched out a model of the neural basis underlying the attentional blink.
With a view to investigating this proposal through modelling, I participated
in the development of NeMo, a platform for the real-time Neural Modelling of
spiking neurons using GPUs. If time permits, I will also present the award
winning project I led at the Barcelona Cognition, Brain & Technology summer
school this year, investigating emotional conditioning of the humanoid robot iCub.


Roesch,  E. B.,  Sander,  D.,  and Scherer,  K. R. (2007) The link between
temporal attention and emotion: A playground for psychology,  neuroscience,
 and plausible artificial neural networks. In J. Marques de Sa et al. (eds.),
Proceedings of the International Conference on Artificial Neural Networks,  volume 2; pp 859-868.

Fidjeland,  A.,  Roesch,  E. B.,  Shanahan,  M. P.,  Luk,  W. (2009). NeMo:
A platform for Neural Modelling of spiking neurons using GPUs. 20th IEEE
International Conference on? Application-specific Systems,  Architectures and Processors.

Roesch,  E. B.,  Sander,  D.,  Scherer,  K. R. (in press). Emotion and
motion in facial expressions modulate the attentional blink. Perception.
16 Dec (4.15pm) 1W 2.7 Hagen Lehmann Examples for the application of agent based modelling in behavioural ecology

In this talk a set of models will be presented which test the effects of different
variables on the evolution of social dominance and social hierarchy in gregarious
animals. These variables are on the one hand environmental pressures like predation
and different rates of food availability, and on the other selective mechanisms like female
mate choice and matrilineal dominance inheritance. The talk will also give an overview
of the potentials and limits of the usage of agent based models in the field of behavioural ecology.
16 Dec 1W 2.7 Jan Drugowitsch Decision Making in the Brain

It has recently become clear that humans and animals feature statistically near-optimal performance in common low-level tasks, and also show use of statistical information in various high-level tasks. This implies that the brain is able to perform the computations that underlie statistical inference, but it remains unclear exactly how and where this computation is performed. The aim of this talk is to give a broad overview over some areas where statistical optimality is observed, and to provide a more detailed discussion about investigating optimality of behaviour in tasks that require decision making under time constraints. In such tasks the decision maker needs to, firstly, accumulate uncertain information, and to, secondly, make a decision once sufficient evidence is available. Regarding the accumulating stage I will discuss the optimal strategy, neuronal measurements and computational models that attempt to explain these measurements. The problem of when to commit to a decision will be addressed purely from the optimality point-of-view, by describing the optimal strategy and comparing it to human and animal behaviour. I will present both standard approaches available in the literature and recently published and yet-to-be-published work performed by myself and others in my lab.
18 Dec (2.15pm) 1E 3.6 Ken Kahn The Modelling4All Project: A web-based modelling tool embedded in Web 2.0


AI at Bath