Speaker
Adriano Barra
Description
In this talk, keeping a statistical mechanical perspective, I will first
revise the link between learning in Restricted Boltzmann Machines and
retrieval in Hopfield Neural Networks. Then, I will focus on the
importance in the choice of the nature of both neurons and synapses
(e.g. digital vs analog) for both learning and retrieval mechanisms.
Finally, I will discuss some variants of the Hebbian paradigm that allow
the network to saturate the critical storage capacity, reaching
Gardner's bound for symmetric networks.
Minimal Reference:
[1] Barra, A., Genovese, G., Sollich, P., & Tantari, D. (2018). Phase
diagram of restricted Boltzmann machines and generalized Hopfield
networks with arbitrary priors. Physical Review E, 97(2), 022310.
[2] Fachechi, A., Agliari, E., & Barra, A. (2019). Dreaming neural
networks: forgetting spurious memories and reinforcing pure ones.
Neural Networks.
[3] Agliari, E., Barra, A., Galluzzi, A., Guerra, F., & Moauro, F. (2012).
Multitasking associative networks. Physical review letters, 109(26),
268101.
Primary author
Adriano Barra