In current implementation, the layers in a neural network can be connected in a line. However, state-of-the-art CNN models such as GoogleNet, layers are connected in computation graph - a directed acyclic graph.
Thus, like other deep neural network frameworks such as Caffe, dolphin-dnn should train neural network models that consist of layers connected in a DAG.
In current implementation, the layers in a neural network can be connected in a line. However, state-of-the-art CNN models such as GoogleNet, layers are connected in computation graph - a directed acyclic graph. Thus, like other deep neural network frameworks such as Caffe,
dolphin-dnnshould train neural network models that consist of layers connected in a DAG.