Establishing tight links between cognitive function and neural activity is a fundamental goal of systems neuroscience. Recent technological developments, particularly in statistical learning techniques and scalable computing, have renewed emphasis on quantitative modeling approaches that relate brain activity to perception, cognition and action. Such quantitative models allow for evaluation of neural representation and, importantly, for prediction. This opens new vistas on brain functions and the possibility to link neural activity with cognition in a way that is similar to approaches in other sciences, e.g. in physics.
Recent development of new robust statistical learning techniques make possible to extract information from modern high dimensional brain recordings even when only limited amounts of data are available. Two related approaches have been developed over the past decades to model neural representation -- encoding and decoding models. The main difference between these models is whether brain activity is predicted from e.g. stimuli (encoding) or whether subjective/cognitive states are predicted from brain activity (decoding). The decoding approach has been very successful in the context of brain computer interface, for example in technical applications and to reconstruct movement, visual, or auditory stimuli from recorded brain activity. Similar methods can be used for the inverse goal, to predict brain activity given an observation of the world, a visual or auditory stimulus or behavior. The encoding approach describes how the brain transforms information in the world into a neural representation. Such approaches have been originally developed in single cell physiology, e.g. to derive receptive or movement fields, and their use was largely restricted due to the lack of robust model estimation techniques and limited computer power. Both, encoding and decoding models are now increasingly used as novel research tools to derive data driven hypotheses about brain networks underlying cognitive function and to show tight coupling between brain activity and perception, even on single trial basis.
This Frontiers Research Topic provides an overview and examples of current work using data driven statistical learning techniques to derive models of neural function and to establish tight links between subjective experience and brain activity. In addition, we welcome conceptual papers that critically assess promises and limitations of data driven statistical learning approaches in neuroscience.
Establishing tight links between cognitive function and neural activity is a fundamental goal of systems neuroscience. Recent technological developments, particularly in statistical learning techniques and scalable computing, have renewed emphasis on quantitative modeling approaches that relate brain activity to perception, cognition and action. Such quantitative models allow for evaluation of neural representation and, importantly, for prediction. This opens new vistas on brain functions and the possibility to link neural activity with cognition in a way that is similar to approaches in other sciences, e.g. in physics.
Recent development of new robust statistical learning techniques make possible to extract information from modern high dimensional brain recordings even when only limited amounts of data are available. Two related approaches have been developed over the past decades to model neural representation -- encoding and decoding models. The main difference between these models is whether brain activity is predicted from e.g. stimuli (encoding) or whether subjective/cognitive states are predicted from brain activity (decoding). The decoding approach has been very successful in the context of brain computer interface, for example in technical applications and to reconstruct movement, visual, or auditory stimuli from recorded brain activity. Similar methods can be used for the inverse goal, to predict brain activity given an observation of the world, a visual or auditory stimulus or behavior. The encoding approach describes how the brain transforms information in the world into a neural representation. Such approaches have been originally developed in single cell physiology, e.g. to derive receptive or movement fields, and their use was largely restricted due to the lack of robust model estimation techniques and limited computer power. Both, encoding and decoding models are now increasingly used as novel research tools to derive data driven hypotheses about brain networks underlying cognitive function and to show tight coupling between brain activity and perception, even on single trial basis.
This Frontiers Research Topic provides an overview and examples of current work using data driven statistical learning techniques to derive models of neural function and to establish tight links between subjective experience and brain activity. In addition, we welcome conceptual papers that critically assess promises and limitations of data driven statistical learning approaches in neuroscience.