feedforward network
A neural network where computation flows in one direction from input to output, layer by layer, without feedback to earlier layers.
As defined in 336 | Anil Ananthaswamy on the Mathematics of Neural Nets and AI
Where it comes up · 1
Most important claims · 2 of 2Search all claims for “feedforward network” →
- Using a trained network (e.g., ChatGPT inference) is purely a feedforward process where an input vector propagates layer by layer to an output, whereas training requires the back-and-forth of computing a loss on the output side and backpropagating it to update each layer's weights toward the input side.336 | Anil Ananthaswamy on the Mathematics of Neural Nets and AI
- Hopfield networks are fully connected recurrent neural networks where neuron outputs feed back into the rest of the network, which differs fundamentally from today's feedforward networks in which computation proceeds from input to output without feedback.336 | Anil Ananthaswamy on the Mathematics of Neural Nets and AI