The ability to adapt to changes in the environment and to optimize performance against undesirable stimuli is among the hallmarks of the brain function. Capturing the adaptivity and robustness of brain function in real-time is crucial not only for deciphering its underlying mechanisms, but also for designing neural prostheses and brain-computer interface devices with adaptive and robust performance. Thanks to the advances in neural data acquisition technology, the process of data collection has been substantially facilitated, resulting in abundant pools of high-dimensional, dynamic, and complex data under various modalities and conditions from the nervous systems of animals and humans. The current modeling paradigm and estimation algorithms, however, face challenges in processing these data due to their ever-growing dimensions. This research addresses these challenges by providing a unified framework to efficiently utilize the abundant pools of data in order to deliver game-changing applications in systems neuroscience.
Converging lines of evidence in theoretical and experimental neuroscience suggest that brain activity is a distributed high-dimensional spatiotemporal process emerging from sparse dynamic structures. From a computational perspective sparsity is a key ingredient in rejecting interfering signals and achieving robustness in neural computation and information representation in the brain. The main objective therefore is to develop a mathematically principled methodology that captures the dynamicity and sparsity of neural data in a scalable fashion with high accuracy. By focusing on the auditory system as a quintessential instance of sophisticated brain function, this research investigates several fundamental questions in systems neuroscience such as plasticity, attention, and stimulus decoding. The research is integrated with education and outreach activities including high school level hands-on workshops, undergraduate capstone projects, and interdisciplinary course development.
