Core Concepts
GCUs bridge biological neurons to gated RNNs with time gates, offering competitive alternatives.
Stats
"We introduce Gated Chemical Units (GCUs), which establish the formal connection between biological-neuron models and gated RNNs, for the first time."
"The GCU-STG performed best, achieving an accuracy of 84.99% being slightly better than the GCU-ATG."
"The GCU-ATG had the best performance, by achieving an accuracy of 87%."
Quotes
"We introduce GCUs, which constitute the first formal derivation of a gated RNN from saturated EECs."
"GCUs perform as well or better than their gated recurrent counterparts on a comprehensive set of tasks used to assess gated RNNs."