Depression & Anxiety Computational Neuroscience Lab

Yujia Peng

School of Psychological and Cognitive Sciences

Peking University

彭玉佳，抑郁与焦虑计算神经实验室

心理与认知科学学院，北京大学

Yujia Peng

Assistant Professor of Psychology

Peking University

DACN lab

The Depression and Anxiety Computational Neuroscience (DACN) lab is affiliated with the Department of Psychological and Cognitive Sciences, Peking University, Beijing. We closely collaborate with the Institute for Artificial Intelligence, Peking University, and the Beijing Institute for General Artificial Intelligence (BIGAI). DACN lab overall focuses on the multi-dimensional mechanistic investigations of mood and anxiety disorders, with a special focus on computational psychiatry and neuroimaging. We use a combination of behavioral experiments, fMRI, EEG, MEG, computational modeling, and machine learning, to promote our understanding of mechanisms underlying mental disorders.

Cognition, emotion, and social processes closely intertwine, and dysregulated functioning of corresponding neural networks associates with mental disorders such as depression and anxiety. Our research examines the order, disorder, and interconnections of cognitive, emotional, and social processing throughout the lifespan from childhood to older adulthood in healthy and diseased brains, to develop improved diagnostic and prognostic tests to be used in community mental health settings. Our current works aim to (1) understand negative cognitive bias in social anxiety, (2) decode individual differences in emotion and social perception based on multi-dimensional data, and (3) develop personalized neurofeedback treatments combined with classic psychological interventions to anxiety through computational neuroimaging methods.

Interests

Depression and Anxiety
Social Anxiety
Social cognition
Action recognition
Artificial Intelligence

Approaches

Computational Psychiatry
Neuroimaging
Computational modeling
Artificial Intelligence

Research Projects

Emotion processing and intention inference in social anxiety disorder

Human interaction with people and society is essential for a normal human life. Social anxiety disorder is an important subtype of anxiety disorders which manifests as extreme anxiety and avoidance of social behaviors and situations. Social anxiety has a high degree of comorbidity with depression and anxiety disorders, with patients showing abnormalities in emotional processing and social cognition, and is an effective entry point for analyzing the comorbidity and heterogeneity of depression and anxiety disorders. However, much remains unknown about the understanding of cognitive and brain mechanisms of social anxiety. We aim to combine psychophysics, eye-movement capture, brain imaging, and physiological signal recordings to investigate the processing specificity of social anxiety patients for emotional and social information in human actions. We aim to reveal individual differences and construct predictive models of social anxiety symptoms. We also aim to explore combinations of different treatments and interventions to promote precise psychiatry.

Neurobiological substrates of mental disorders and potential treatments

The field of psychiatry faces the dual challenges of heterogeneity and co-morbidity, in terms of mental disorder manifestation and pathological mechanisms. Currently, our understanding of the neural mechanisms of mental disorders is very limited, and both diagnosis and treatment basically remain at the level of relying on the subjective symptom reports of the DSM-5. It is widely recognized that the field has not yet established objective and effective diagnostic criteria and intervention and treatment programs. To address these challenges, we aim to integrate interdisciplinary approaches, using brain imaging, psychophysics, physiological signal recording, advanced statistical and computational modeling, and machine learning to explore in-depth the cognitive neural mechanisms, developmental patterns, and innovative treatments of depression and anxiety disorders.

The hierarchical representation of human actions and convolutional neural networks

The human visual system can process external visual motion information quickly and efficiently, and obtain a series of understandings from low-level information (e.g. speed, texture) to high-level information (e.g. intention, goal). Action recognition is mediated by complex multi-stage processing of visual information emerging rapidly in a distributed network of cortical regions in the ventral and dorsal pathways. Previous evidence showed that image information and motion information are mainly processed in the ventral and dorsal visual pathways respectively. However, the complicated process remains poorly understood and is hard to be captured by computational models. With the development of deep learning and neural networks, computers have reached a height close to the human level in many visual tasks. But computers rely on a large amount of training data and strictly defined computational rules, which is not close to the highly efficient and adaptable human visual system. To investigate the representation of human actions in human brains and artificial neural networks, we are currently using fMRI and MEG to investigate the neural mechanisms behind action understanding and the correspondence between neural networks and human neural systems.

Perceiving causal human actions

We often have a strong sense of causality as actions unfold. Human actions and interactions can be interpreted in a causal framework. First, the actions are caused by goal and environment, and when there are human-object interactions, the status change of the object is caused by the actions and objects. I aim to examine the role of causality in the perception of human actions from different aspects. 1) whether human observers have a causal expectation of body movements and 2) whether the causal expectation influences the perception of actions..