Dynamic Visual Inferences and their Neural implementation – DyVINE

Theoretical framework: Our goal is to clarify the fundamental nature of neural processes in the brain that give rise to intelligent perception, cognition and action through a close integration of compatible theoretical, behavioral and neurophysiological approaches.

Hypotheses and objectives: A customary framework of such an investigation relies on signal-detection theory, the behavioral task of simple decision making and neural measurements of either single-cell recording or fMRI. In contrast, our approach is based on three novel hypotheses: a) the proper theoretical framework for capturing brain processes is probabilistic inference, b) studies of simple decision making overlooks the full complexity of behavior that strongly depends on the brain’s internal representation of the situation, and c) meso-scale dynamic analysis of the corresponding neural signals is essential for understanding the underlying circuit mechanisms and, by itself, neither single-cell recordings nor fMRI can support this requirement. Based on these hypotheses, our objective is to develop an adequate integrated framework that addresses the above shortcomings and to use the framework to provide neural evidence for probabilistic computation in the brain.

Approach: We combine human and animal behavioral and eye movement measures in complex decision making tasks with optical imaging in the primary visual cortex of the behaving non-human primate (NHP) to understand the process of “explaining away”, one of the fundamental features of probabilistic inference.

Innovation: This project goes beyond the state-of-the-art in the field on multiple counts. First, our behavioral paradigm for both humans and NHPs will use a novel complex decision making task with multiple equally possible explanations of the sensory input in any given trial, and we will use this paradigm to identify, based on behavior, the internal representation the observers use in their decision. Second, for neural measurements, we will use dynamic stimuli and the methodology of optical imaging with voltage sensitive dyes thereby combining the advantages of capturing neural activity changes at the temporal resolution of milliseconds with large-scale measurement of neural population at the same time. This allows real-time inspection of the decision process at the neural level. Third, instead of using the typical tool of tracking uncertainty through the variance of the neural signal, we will tie dynamic patterns emerging in the brain to behaviorally identified latent variables to directly validate the explaining-away process in early cortical representations. The outcome of our project will not only include strong evidence supporting the recently emerging new view on probabilistic computation in the brain, but also a set of novel tools for further rigorous investigations.

Primary researchers involved: Anna Montagnini and Frédéric Chavane from Institut de Neurosciences de la Timone, Marseille and József Fiser from Central European University, Vienna.