Human and Neuro-Inspired AI Visual Reasoning
Different brain areas are segmented with functional localizers and analyzed to assess the representation of scene meaning.
Deep Neural Networks (DNNs) have led to tremendous progress in the ability of AI to search and recognize objects, linguistically describe scenes, and navigate. Yet, current DNN-based AI approaches fall short of matching humans in reasoning capabilities, robustness to dynamic changes in the environment, and learning from few samples. With this project we will investigate how humans learn to visually reason about complex scenes, how the brain represents functional/semantic knowledge structures, and the requirements to train humans to acquire such knowledge structures. We will then use the gained knowledge about humans to create new AI solutions so that machines can acquire higher-level understanding and visual reasoning about scenes.
This project will advance our understanding of how the human brain represents and learns to attain high visual reasoning abilities and develop new AI methods that can improve the ability to make complex inferences from scenes and improve visual-language navigation.
