Using Artificial Intelligence (AI) to Understand Neural and Behavioral Variability
PI: Xiao Liu (Biomedical Engineering)
The brain can exhibit distinct responses to identical sensory input or task performance at different times. For example, the same visual stimulus may evoke neural responses of varying amplitude in the visual cortex; and the same person may have distinct reaction times to a task in different experimental trials. This neural/behavioral variability is increasingly recognized as a meaningful feature of brain function, reflecting differences in attention, perception, internal states, and individual traits. It also varies across different age groups and is altered in brain diseases, such as Alzheimer’s disease (AD) and autism. Understanding the neural variability can provide insights into how the brain processes information under dynamic conditions, improve brain-computer interface by accounting for these fluctuations, and reveal early biomarkers for neurological or psychiatric disorders based on it. However, the underlying neural basis of this neural variability remains largely unknown. While artificial neural networks (ANNs) were originally inspired by brain science, they are now being used to help understand brain functions and mechanisms. Recent studies have demonstrated similarities between ANNs and biological brain networks and how this similarity can account for performance differences between ANN models. These findings prompt the question of whether the neural variability can be presented and understood as changes of neural similarity to ANNs, and whether this brain-ANN similarity explains behavioral differences. The goal of this project is to quantify the neural variability from the perspective of brainANN similarity and then examine whether this similarity explains variability in behavior. Our hypothesis is that the brain’s response to visual stimuli varies over time in its similarity to ANN responses, and that greater similarity is associated with better subsequent memory of the stimuli. Specifically, we will compare the activation patterns of biological brain networks and ANNs to identical visual stimuli and then examine how the variation of their similarity predict differences in memory outcomes.
List of specific areas of expertise or skills:
The student is expected to have some experience with ANNs, neural data (fMRI, EEG, neuronal recordings, etc.), and programming in Matlab and/or Python. She/he needs to be available for weekly research meetings to report progress and discuss the project.
List of specific objectives:
The development/ and ptimization of ANNs to decode visual stimulation from neural signals. The use of these models to estimate the brain’s encoding efficiency. The algorithm and metrics for quantifying the similarity between ANNs and brain networks across different hierarchies/layers. All these preliminary data will be used to support a grant proposal to NIH or NSF. The main result will be submitted for publication in a peer-reviewed journal.
Medium to long-term goal:
To secure an external funding to continue and expand research on the relationship between biological and artificial neural networks.
Connection to ICDS’s mission:
We will develop and apply state-of-art AI models to understand brain functions. The project is also to understand the ANN from the perspective of the brain science.
Recent and/or planned engagement with ICDS:
As a ICDS co-hire, the PI has been actively engaged in various ICDS activities, including committee work, symposiums, retreats, and luncheons, and plans to continue this engagement. The junior researcher supported by this program will also actively participate in future ICDS activities involving students. Both the PI and junior researcher will also seek opportunity to present this project at ICDS events, such as ICDS symposiums and/or luncheons.