Energy efficient Bayesian synapses
| Autor: | Malkin, James, O'Donnell, Cian, Houghton, Conor, Aitchison, Laurence |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2022 |
| Předmět: | |
| DOI: | 10.5281/zenodo.7071494 |
| Popis: | Recent work (Aitchison et al. 2021) proposes probabilistic synaptic sampling; moreover, that synaptic plasticity performs Bayesian inference. In this paradigm, the settings of synaptic weights are parameters to be inferred from noisy sensory input and feedback. Model uncertainty may be expressed via variation in synaptic weights. However, stochastic synapses invoke unreliability which may be detrimental to performance, yet unreliability is a hallmark of biologically synapses. We consider an evolutionary imperative to balance performance with energy-use. The proposed trade-off casts energetic efficiency as a constraint on synaptic physiology. We consider the physiological costs of reliable transmission; and whether the resulting energy efficient learning objective may induce probabilistic synapses that mirror the Bayesian posterior. {"references":["Aitchison, L., Jegminat, J., Menendez, J.A. et al. Synaptic plasticity as Bayesian inference. Nat Neurosci 24, 565–571 (2021). https://doi.org/10.1038/s41593-021-00809-5"]} |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|