Towards an Optimal Control Perspective of ResNet Training
This work addresses training inefficiencies in deep neural networks, offering a theory-grounded approach for pruning, though it appears incremental as it builds on existing ResNet architectures.
The authors tackled the problem of training ResNets by formulating it as an optimal control problem, resulting in a method that biases unnecessary deeper layers to vanish, indicating potential for layer pruning.
We propose a training formulation for ResNets reflecting an optimal control problem that is applicable for standard architectures and general loss functions. We suggest bridging both worlds via penalizing intermediate outputs of hidden states corresponding to stage cost terms in optimal control. For standard ResNets, we obtain intermediate outputs by propagating the state through the subsequent skip connections and the output layer. We demonstrate that our training dynamic biases the weights of the unnecessary deeper residual layers to vanish. This indicates the potential for a theory-grounded layer pruning strategy.