Themes
Models of Neural Representations: How does the brain understand the world around us? How does the brain control the body? Or asked more abstractly: How do complex patterns of neural activity relate to the environment in which we live? The old problem of Brain Representations - their nature and how to understand them - has come back to life in recent years, as our methods to record brain activity have dramatically improved. Like never before in history, we now have enough data to potentially crack the brain’s secret code. Our group is developing cutting-edge mathematical and statistical tools to build and test representational models and applies these techniques to understand a broad range of mental functions.
Visual Processing in Man and Machines: Artificial Neural Networks are bio-inspired artificial intelligence (AI) systems in which information passes between layers in a similar way to neurons in the cortical sheets of the brain. Despite the recent success of deep neural networks (DNN) at recognizing visual images, a lot more progress needs to be made before systems understand the visual world as humans do. We study the brain information-processing strategies and emulate them in the architecture and objective functions of DNN models. With this line of research, we aim to develop more human-like AI systems.
Machine Learning for Neuroimaging Data: We develop new machine learning methods for neuroimaging data analysis. We use supervised and unsupervised learning techniques, as well as deep neural networks to model high-dimensional brain signals (e.g., functional MRI, EEG), and discover the hidden structures in the data. Specifically, we aim to develop novel multivariate analysis methods:
- to evaluate the generality and validity of information and structures across subjects,
- to fuse brain data across modalities for the analysis of brain connectivity, and
- to decode both spatial information and temporal information in a unified framework
