Your Brain Segments Images Like a Detective, New Study Finds
This could lead to smarter AI that sees the world more like we do.
Deep Dive
Neural dynamics in the visual cortex are poorly understood, especially for static stimuli. A biologically plausible model of cortical computation assumes that inferring image features is coupled with inference about how those features are organized into meaningful parts, also known as segmentation.
Given an input image, the model iterates between inferring features and segments through recurrence and feedback, inducing dynamics in the neural representation of features. The model yields theoretical and exact predictions and predicts both gain and
Key Points
- The brain uses a two-step process: first identifying features, then grouping them into meaningful parts.
- A new computer model mimics this process and matches real brain cell recordings.
- This could lead to smarter AI for self-driving cars and medical images.
Why It Matters
Better AI for self-driving cars and medical imaging, and deeper insight into brain disorders.