Astrocyte Biocomputing: Astrocytic Contributions to Neural Network Computation
Summary
- Recent biological studies reveal that astrocytes, which make up about 50 percent of brain cells without emitting electrical signals, play key roles in cognitive functions like learning and memory.
- Researchers propose brain-inspired machine learning algorithms based on two key astrocyte characteristics: modulating thousands of synapses simultaneously and slow, rhythmic non-electrical communication.
- A model mimicking simultaneous synapse modulation yields faster training, better noise robustness, and a supervised learning algorithm applicable to any neural network.
- A rhythmic sharing algorithm based on non-electrical communication yields rapid anomaly detection adaptation and prevents catastrophic forgetting, outperforming neurocentric models.
Why it matters
Incorporating astrocyte-inspired biocomputing concepts into machine learning models can help remediate critical weaknesses of current architectures, such as catastrophic forgetting and slow training.