Decoding the activity of neural networks: biological decoding and constant metric content

Máté Lengyel, P. Érdi, Alessandro TrevesSSRN 2003

Abstract

First, we determined what constraints have to be fulfilled if neurons performing dot-product operations between their inputs and synaptic weight vectors have to achieve quasi-optimal decoding of their inputs, the optimum being defined by Bayesian decoding. We showed that Hebbian plasticity may not provide neurons with the neccessary weight vectors for this purpose. Second, we investigated why metric content of neuronal population responses was found to be constant regardless of how many neurons were used for decoding. We found that homogeneity of tuning precision among neurons account for this, and the contribution of neural oscillations and network architecture may also be signigificant.
BibTeX
@misc{lengyel2003decoding,
  author = {Máté Lengyel and P. Érdi and Alessandro Treves},
  title = {{Decoding the activity of neural networks: biological decoding and constant metric content}},
  year = {2003},
  howpublished = {SSRN},
}

Loading related content...

Figure