We humans, and all living creatures, must interact with the world around us throughout our life. Fortunately, our nervous systems allow us to gather sensory information, process it and act upon. How these processes occur still remains one of the most captivating questions in neuroscience. The answer not only has far reaching consequences to medicine, artificial intelligence, and other scientific and technology fields. They may also profoundly impact economy, society and overall human well-being.
Sensory neurons have long been shown to transform specific physical features of the outside world (e.g. colours, contrasts, and sounds) into sequences of electrical impulses or spikes. This process is called encoding and generates spike sequences that are both noisy and correlated. Noisy, because they may vary even after repeated presentations of the same stimulus. Correlated, because they may depend on the sequences generated by other neurons. These observations continue to raise very interesting and heatedly debated questions about what aspects of the neural responses are informative and what sort of algorithms should the brain or a prosthetics employ to efficiently learn that information.
To answer these questions, I have analysed recordings of neural activity ranging from single neurons to the whole human brain using computational and statistical methods. However, I soon realised that oftentimes different methods yield seemingly contradictory conclusions. Most importantly, I found out that their otherwise eloquent narrative and conjectures are not always justified by their underlying quantitative framework. Turns out that this is not something new. Much research in neuroscience has been devoted to clarify the very complicated subtleties of otherwise deceivingly simple methods and the questions they attempt to address. Yet, this research has not unveil all potential contradictions and inconsistencies of classical methods, let alone those introduced by new ones.
My research has mostly been devoted to fill this gap, exploring the mathematical properties of the methods currently employed in the field of neural coding. My main tools stem from the fields of information theory, statistics, dynamical systems, and machine learning, which I employ to separate facts from misconceptions through rigorous mathematical proofs and computations. I believe that, by revealing and resolving their conceptual ambiguities and mathematical limitations, my research will potentially and eventually contribute to a more general and well-grounded understanding on how the nervous system works and what can be done to improve life quality and well-being for those dealing or recovering from nervous system damages or disorders.