New AI Learns Like a Brain, Could Make Computers Smarter
This could lead to AI that thinks more like us—and uses less energy.
Predictive Brain Networks (PBNs) are a new hierarchical recurrent framework that uses local predictive-coding-based learning. According to the article, PBNs achieve representational alignment with the human ventral visual stream matching or exceeding established vision models, and their top-down predictions encode unique brain-aligned information not captured during bottom-up feature extraction. PBNs also outperform existing predictive-coding-inspired networks on downstream classification, reproduce key predictive-coding signatures seen in primate cortex, and support uncertainty estimation—combining brain-inspired learning with the representational power of modern deep networks.
- Predictive Brain Networks (PBNs) learn like the brain, using local predictions instead of global error signals.
- PBNs match or beat standard AI in mimicking brain activity and classifying images, and they can estimate uncertainty.
- This could lead to more energy-efficient AI that runs on smaller devices and learns continuously.
Why It Matters
More efficient AI could mean smarter phones, better robots, and lower energy bills for tech companies.