The NeuroComputation and Biological Vision Team (NeuroBiT) conducts fundamental, experimental and computational modelling research on the mechanisms of human visual perception, being at the frontier between Computational Neuroscience and Computer Vision. It addresses problems by combining both, computational modelling tools and psychophysical experimentation. Recent research lines of our members, have been focused on the study of processes such as colour perception (categorisation, chromatic induction and colour constancy), high dynamic range visualisation techniques; visual saliency, and visual aesthetics. Most of our past research experience and background involves the development of high-level biologically-inspired computational models whose fundamentals are firmly grounded on psychophysical studies of human perception. Our present interest also includes the definition of biologically-plausible, mid-level architectures which simulate the network dynamic characteristics of the human visual cortex and the use of computational learning techniques to map the output of our models to complex phenomena. Our final objective is to apply our knowledge to solve important Computer Vision problems and provide some answers to medical and biological questions.
One of the present projects is the development of spiking models for the studying of several perceptual processes in the human visual system. Concretely, we study the formation of the neuronal connections in the cortex and the role of neural oscillations in the learning process. This study is performed using computational spiking models using Spike Timming Dependent Plasticity, a type of plasticity present in biological systems that are the neuronal basis for learning processes. These architectures can only be implemented in computational clusters or supercomputers and generate a huge amount of data (around 10 Gb for a single hypercolumn simulation) which requires a specialized software for its statistical study. Nonetheless, the architecture has to be continuously modified (e.g. ablations) for the study of the impact of every computational element into the final system dynamics. One of the important features to be studied is the generation of neuronal oscillations and its role for the formation of network connectivity and its associated learning process.
Main responsibilities:
● Developing software to perform a statistical study of the results generated by an existing computational architecture, and the modification of this architecture.
Deadline: 27/02/2024
Més posts de Recerca
Comentaris