Simon Thorpe Heilige Gralsannäherung? Ultra-Sparse Representations in Neural Networks: Biological Inspiration for Artificial Intelligence? •Deep Learning Networks are the state of the art Problem •Very computationally demanding Inspiration from biology •Use spikes! •Use the order of firing to encode information •Use circuits to control the percentage of neurons that fire •Grandmother Cells and Neocortical Dark Matter •Spikes also important for learning •another story! Development of Spiking Hardware •Ultra Low Power devicesFinal Thoughts Don’t believe the Neuroscientists! •Neurons don’t always use rate coding •They send pulses not floating point numbers •Neurons can be extremely selective •Neocortical Dark Matter may be the reality •Ultra Sparse Coding!! http://dobigeon.perso.enseeiht.fr/SPARS2019/Thorpe_SPARS_201… |
|
aus der Diskussion: | Diskussion zu BrainChip Holdings |
Autor (Datum des Eintrages): | H2OAllergiker (29.07.19 20:06:28) |
Beitrag: | 271 von 15,198 (ID:61128947) |
Alle Angaben ohne Gewähr © wallstreetONLINE |