Human vision receives over 80% of sensory inputs to the brain. The neural and computational mechanisms of visual perception have been a central topic in cognitive neuroscience. My research questions along this line of research include: (1) the computation and representation of uncertainty in perceptual decision-making; (2) the influences of top-down modulation (e.g., attention/learning) on population codes in the human brain; (3) the neural implementations of Bayesian inference; (4) neural mechanisms of visual working memory.
The past few years have seen the surge of comparisons between the artificial and the human visual systems. We see a promising future that research from computer vision and human vision can mutually benefit and foster the development of general intelligence. My research questions include (1) pixel-level image reconstruction from brain activity based on deep networks (2) neural mechanisms of face recognition in natural scenes.
Computational Psychiatry is an interdiscipline that bridges basic computational neuroscience and translational psychiatry. Research in computational psychiatry emphasizes using the computational models established in normal subjects or basic neuroscience to characterize the mechanisms of abnormal cognitive behavior in psychiatric diseases. I am currently working on the following questions: (1) computational mechanisms of visual working memory deficits in patients with psychiatric diseases; (2) abnormal reinforcement learning in psychiatric diseases; (3) reward-based emotional regulation in psychiatric diseases.
Functional magnetic resonance imaging (fMRI) has been widely used in investigating complex structural and functional compositions of the human brain, as well as in various forms of diagnostic and rehabilitative practices. The rapid progress of neuroimaging techniques in the last couple of decades has produced a vast quantity of brain anatomical and functional data. It remains a key challenge for neuroscientists to efficiently infer substrates of human brain functions from massive datasets, and also for biomedical practioners to extract clinically meaningful information. Hence, developing novel data processing methods and pipelines is essential and fruitful for the next stage of neuroimaging research. We believe that bridging state-of-the-art machine learning methods and neuroimaging data will open new avenues for future data-driven neuroscience research and promote more accurate clinical decisions, as well as customized medical services for patients.