(A) A high-level overview of the proposed neuron–astrocyte network. The Transformer block is approximated by a feed-forward network with an astrocyte unit that ensheaths the synapses between the hidden and last layers (matrix H). Data are constantly streamed into the network. (B) During the writing phase the neuron-to-neuron weights are updated using Hebbian learning rule and the neuron-to-astrocyte weights are updated using a presynaptic plasticity rule. During the reading phase, the data are forwarded through the network, and the astrocyte modulates the synaptic weights H. Credit: Proceedings of the National Academy of Sciences (2023). DOI: 10.1073/pnas.2219150120
Artificial neural networks, ubiquitous …
Read more…….
Be the first to comment