Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space.
Mor GevaAvi CaciularuKevin Ro WangYoav GoldbergPublished in: EMNLP (2022)
Keyphrases
- feed forward
- back propagation
- artificial neural networks
- neural nets
- neural network
- biologically plausible
- hidden layer
- recurrent networks
- neural architecture
- concept space
- feed forward neural networks
- recurrent neural networks
- multi layer
- space time
- visual cortex
- activation function
- fault diagnosis
- neuron model
- training algorithm
- multilayer perceptron
- fuzzy logic
- input output
- low dimensional