TITLE: Ecology-based modelling: a novel approach to understanding perception
SPEAKER: Emma Byrne

ABSTRACT:

Sense organs conflate several features of the environment.  Eyes, for instance, receive a signal that is dependent on illumination, transmittance and reflectance of light from the world. The environment will always be a source of ambiguity that can only be minimised by taking context into account.

In order to model perceptual systems, whether to understand their natural counterparts, or to build autonomous artificial intelligences, we require a more complete understanding of how ambiguity is resolved.
Biologically-inspired perceptual modelling has so far focused on existing animal visual systems. These models either mimic known neural architecture or observed behaviour.

Here we propose an alternative approach: ecology-based modelling.
Although bio-inspired approaches are not new, the ecology-based approach is fundamentally different to these existing methods. Rather than imitating aspects of a particular animal’s physiology or behaviour, the novelty of this approach is to focus on the ecologies of an agent and its process of adaptation. These features can be observed and measured and can then act as inputs to an ecologically drive process of evolution and development.

There are several advantages to this approach, compared to traditional bio-inspired methods, in addressing general questions of ambiguity resolution. Firstly, natural systems have taken specific routes to resolving ambiguity; therefore the modelling of any one system prejudges the answer to a large extent.  However, once a naturalistic ecology has been defined, multiple agents can adapt to it, and their common features can be extracted. Secondly, each natural perceptual system studied represents a single point in time in the evolution of perception. An ecology-based approach allows us to study the evolutionary processes that drive increasingly complex perceptual systems in response to increasingly complex challenges to survival. Here, we discuss this ecology-based approach in more detail, develop a formal framework for the approach, and use it to investigate the perceptual spaces of human neural processing.