A proposal to train each neuron independently on a hand-assigned data subset, with the resulting specialization presented as a new interpretability and scalability framework, but with no empirical validation.
Bengio, Reinforcement Learning for Deep Neural Architectures: Con- ditional Computation with Stochastic Computation Policies
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.NE 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Switch-Based Multi-Part Neural Network
A proposal to train each neuron independently on a hand-assigned data subset, with the resulting specialization presented as a new interpretability and scalability framework, but with no empirical validation.