Modelling Mental Rotation in the Brain Using Deep Learning

Explore the fascinating intersection of neuroscience and deep learning in this 49-minute lecture on modelling mental rotation in the brain. Delve into the human ability to mentally manipulate 2D and 3D object representations, a skill that deep networks currently lack. Examine ongoing research efforts to understand brain processes during mental rotation and how similar operations could be implemented in deep learning systems. Learn about the differences between the human visual system and deep networks, particularly in recognizing objects from unusual viewpoints. Discover insights from fMRI data, visual transformers, and comparative studies between humans and AI models. Gain valuable knowledge about representations in the brain, the visual system, and potential future developments in this field from Stephane Deny, an assistant professor at Aalto University with expertise in computational neuroscience and machine learning.

Introduction
Representations in the brain
The visual system
Visual system vs deep networks
Mental rotation
Natural objects
Question
Visual Transformer
Results
Failure mode
Results with humans
What happens in the brain
The model
Test results
Equivalence
Motivation
FMRI data
Questions
Recurrence
Future work