Beomyeol Jeon

In the current DNN implementation, the weights of layers such as fully connected layers and convolutional layers are initialized with random values from a Gaussian distribution. It is known that...

Load pre-trained neural network models and save trained models

1

This issue is brought from cmssnu/dolphin#104. Current DNN does not support loading and saving neural network model parameters. Since it usually takes several hours or even several days to train...

Add the neural network configuration file for GoogLeNet

1

GoogLeNet, described in [Going Deeper with Convolutions](http://arxiv.org/abs/1409.4842), is a 22 layer deep convolutional neural network that won in ILSVRC14 classification and detection. Like AlexNet, GoogLeNet also has been widely used...

In DNN module, each worker outputs the statistics about training and cross validation results such as error rate and the number of instances for partial input data that are assigned...

In the current implementation, the layers in a neural network can be connected in a linear fashion, which means that one layer is only connected to its previous layer and...

`ILPSolverOptimizer` generates transfer steps strictly following the optimization result. However, this may cause many transfer steps with small amount of data. For example, Task A needs to send 1500 data...