Learning and Decision Making
To interact with the world, our brain often relies on a cognitive map, or model, of how objects are organized in the environments, how they may interact, and which scenarios are more or less similar to each other. My colleagues and I have studied the role of one brain region called orbitofrontal cortex (OFC) in representing the brain's cognitive map.
One future research theme of my lab is to understand the computational strategy by which the brain learns to build a cognitive map or generative model of a new environment. The questions include: how do we generate hypotheses about the latent rules in an environment to obtain the desired outcome, and how do we test the hypotheses through trial and error? How do we optimize the experience that can be stored in a limited pool of memory in order to learn about an environment? What form of prior assumptions do we make about a new environment when we first interact with it? How does the OFC's representation of an environment evolve throughout learning?
With the advancement of non-invasive human neural imaging techniques such as functional magnetic resonance imaging (fMRI), neuroscientists now enjoy an ever-increasing amount of open data, better spatial and temporal resolution, and new ways of imaging the brain. But the development of computational tools to analyze these data have not kept pace with these advances.
I have identified and analyzed the source of a statistical bias in a powerful analysis tool called representational similarity analysis, and contributed a Bayesian algorithm that estimate the similarity structure of neural representation with much less bias. Recently, I have developed a new Bayesian tool that optimally extracts the true neurally related BOLD activity underlying the noisy fMRI data, using an imaging method called multi-echo fMRI.
The lab will continue to develop advanced computational tools for processing and understanding neuroimaging data based on Bayesian principle and tools from deep learning. I am also interested in developing novel biomarkers for psychiatric conditions.
Time plays an important role in language understanding, motor control, and the sense of agency. Yet little is understood of how the brain tells the passage of time. I investigate the mechanism of various aspect of time perception. This includes: how we perceive the order between two events and how the brain recalibrate itself to changes in the statistics of timing relation in the environment; what mechanism causes illusion in time perception, for example, why do we perceive a novel stimulus as lasting longer than a repeated stimulus; and how the brain makes inference of duration by integrating various types of information, including the statistical (prior) distribution of the duration of similar stimuli, the (possibly) conflicting cues of duration arising from multiple sensory stimuli, and how such information interacts with short term memory that decays over time.
To study these phenomena, I develop probabilistic models to analyze every single choice made by human participants during experiments and quantitatively compare various models. Such approach takes into account of the encoding and inference process happening in the brain while considering the uncertainty about such processes from the researcher's perspective, and will be the major research angle taken by the lab to understand human behavior.
Brain-inspired artificial intelligence
This line of research will take inspiration from findings in cognitive science and neuroscience to develope new architecture and new learning objective for deep neural networks. Stay tuned!