Evolving to be flexible - optimizing task-dependent information processing in the visual systemUdo Ernst - University of Bremen
Andreas Kreiter - University of Bremen
“Higher animals” typically evolved in complex environments provide ecological niches for species with a wide variety of cognitive abilities and different behaviors. The enormous complexity and variability of this environment suggests a corresponding large variety of computational demands in neural processing. Therefore optimizing dedicated individual neuronal pathways and circuits for each individual task must quickly have come to a limit set by the amount of brain tissue which can reasonably be supported by a given species. One evolutionary principle and way out of this dilemma may have been development of increasing amounts of flexibility in using the same set of neurons and neuronal circuits.Conceptually, this perspective implies co-evolution of function and flexibility: Evolutionary principles not only optimize neural processing towards specific functions, but also shape networks to be able to efficiently and rapidly change between different functional configurations for different behavioral needs.Here we will explore the hypothesis that neural systems optimize their ability of being flexible, and that corresponding networks are constantly adapted on all time scales to maximize flexibility while robustly performing varying tasks. This problem is non-trivial since optimizing for a particular computation and for flexibly switching between different computations naturally compete. Moreover, information processing in the brain involves multiple players engaged in different computations, hence coordination has to be performed in parallel, by partitioning a given task into suitable control inputs to the various players. How can such a parallel coordination strategy evolve from optimizing networks to be flexible? Which network configurations are optimal for flexible computation, and how do they evolve under given physiological constraints? Do multiple solutions for the same flexibility problem co-exist, and can they explain the observed variability of neural responses?To address these fundamental questions, we will investigate optimization of flexibility in theory and experiment, with focus on the visual system. Theoretical work will formally study central computational problems such as parallel control and emergence of circuits proliferating flexibility in the framework of generative models, where co-evolution of flexibility and function can be seen as a constrained optimization problem. Joint modelling and experimentation will focus on selective attention as one central aspect of flexibility in vision involving task-dependent coordination of multiple visual areas.We expect our experiments to reveal features of a dynamics optimized for flexibility, and to exhibit signatures of on-going optimization when subjected to changes in task demands. Biophysically realistic modelling will accompany experiments for testing predictions from formal theory, and for understanding flexibility and optimization mechanisms underlying the recorded data.