The scientific objectives of the research and training program fits those defined in FACETS-ITN and particularly the proposed biological studies which aim at "providing benchmark paradigms for the study of sensory cortical mechanisms involved in low-level (non-attentive) perception.

input at the expense of detailed processing of other visual input. This focal processing may take the form of biased processing towards certain locations or features in the scene. At the same time, outside of the demands of a particular visual task definition, it is important to be alerted to content that may be of interest in its own right, for example a predator suddenly appearing while an animal is searching for food. These two elements constitute respectively, the task driven top-down side of attention which serves to instigate bias towards task relevant content, and the bottom-up side which may be viewed as a stimulus driven component which results in the deployment of attention towards conspicuous visual patterns. Given the key role of visual search in our daily life, it is easy imagine that applying similar strategies for artificial vision systems will give them new major capabilities to interpret more efficiently the incoming flux of images. Following this goal, a variety of models of saliency and attentional bias have recently emerged, coming from both the computational neuroscience or computer vision communities.

In this PhD, the candidate will first focus on the bottom-up side and to investigate a possible general principle for attention. The starting point will be to revisit some recent contribution where attention has been explained as being an emergent property of a neural population. In the literature, this has been modeled by neural fields, more precisely by an integro-differential equation with a mexican-hat-type connectivity between populations. This model is giving a high saliency to positions "x" having an activity "sufficiently different" with respect to a given context, where the context is here geometrically defined by a region surrounding "x". Experimentally, they showed the advantages of such a model, for example in term of robustness. Meanwhile, from a mathematical point of view, much progress has been achieved recently, including from our team at INRIA. So, the goal will be to understand the key features of the model, by applying the powerful tools of bifurcation analysis and numerical continuation to study the changes to the model's solution structure under the variation of parameters. Then, the candidate will further investigate the definition of the context on which the computation of salience is based. The proposal for the scale at which saliency computation takes place varies within the studies we have described, from a local surround region to the entire image, to the space of all natural images. This is a consideration that no doubt impacts upon the resultant judgements of saliency and is deserving of further consideration.