Leveraging Machine Learning to Understand Visual Processing for Interacting with Objects and People
Speaker: Maryam Vaziri-Pashkam, MD, PhD
Living in an environment filled with objects and people, we effortlessly know how to move our bodies to interact with everything around us. The computational challenge lies in dynamically transforming the visual input into the continuous space of possible movements.
In this talk, Dr. Vaziri-Pashkam will present their research that uses machine learning to study the visual processes that enable these interactions.
- behavioral findings showing that when grasping objects, people rely on perceptual features different from those used in object categorization.
- evidence from machine-learning analyses of functional MRI data that reveals robust and abstract representations of objects in the human brain, which may underlie these distinct behaviors.
- behavioral experiments indicating that humans use their knowledge of human body kinematics to predict others’ goals in real time.
- fMRI and machine learning results demonstrating that certain regions in the human visual cortex extract kinematics from observed body movements, suggesting their role in real-time interaction.
By combining computational methods, behavioral findings, and neuroimaging data, Dr. Vaziri-Pashkam's work aims to bridge the gap between the visual input we receive and the continuous motor output we produce.