Partially Trainable Neural Network
A neural network model situated between fixed random feature networks and fully trained networks, in which only some parameters (such as the first-layer neurons or the second layer) are trainable, while the remaining parameters are randomly initialized and held fixed.
Timeline (last 90 days)
A recent paper, through theoretical analysis and experimental verification, revealed phase transition patterns in the approximation bounds and training error of partially trainable neural networks.
In partially trainable neural networks, as the trainable proportion increases, the rate of performance improvement changes significantly: rapid improvement in the initial stage, followed by diminishing returns, with the transition being sharper when the initialization mismatch is more severe.
All Facts (2)
A recent paper, through theoretical analysis and experimental verification, revealed phase transition patterns in the approximation bounds and training error of partially trainable neural networks.
50%UnverifiedIn partially trainable neural networks, as the trainable proportion increases, the rate of performance improvement changes significantly: rapid improvement in the initial stage, followed by diminishing returns, with the transition being sharper when the initialization mismatch is more severe.
50%