Methods and systems for budgeted and simplified training of deep neural networks

A system for a server comprising:

• a processing core having a deep neural network (DNN);
• an I/O controller hub coupled to the processing core and to provide network, data storage, and DNN access for the processing core;
• and
• a graphics processor coupled to the I/O controller hub and to
• train the DNN using a plurality of training sub-images derived from a down-sampled training image, and
• test the trained DNN using a plurality of testing sub-images derived from a down-sampled testing image.

In a recurrent deep Q-network (RDQN) having a local attention mechanism located between a convolutional neural network (CNN) and a long-short time memory (LSTM), a method comprising:

• generating a plurality of feature maps by the CNN from an input image;
• applying hard-attention by the local attention mechanism to the generated plurality of feature maps by selecting a subset of the generated feature maps;
• applying soft-attention by the local attention mechanism to the selected subset of generated feature maps by providing weights to the selected subset of generated feature maps in obtaining weighted feature maps;
• storing the weighted feature maps in the LSTM;
• and
• calculating a Q value for different actions based on the weighted feature maps stored in the LSTM.

A system for a server comprising:

• a processing core having a recurrent deep Q-network (RDQN) including a convolutional neural network (CNN), and long-short time memory (LSTM), wherein a local attention mechanism is located between the CNN and LSTM, and wherein the CNN and LSTM are input and output modules for the local attention mechanism having soft-attention and hard-attention;
• an I/O hub controller coupled to the processing core and to provide network, data storage, LSTM, RDQN, CNN, and local attention mechanism access for the processing core;
• and
• a graphics processor coupled to the I/O hub controller and to
• process an input image and a plurality of feature maps based on the input image,
• apply hard-attention to generated plurality of feature maps by selecting a subset of the generated feature maps,
• apply soft-attention to the selected subset of generated feature maps by providing weights to the selected subset of generated feature maps in obtaining weighted feature maps,
• store the weighted feature maps in the LSTM, and
• calculate a Q value for different actions based on the weighted feature maps stored in the LSTM.