Information processing system having an information processing device and multiple partial reservoirs that are trained, information processing device that trains multiple partial reservoirs, and non-transitory computer readable memory medium that stores information processing program for training multiple partial reservoirs

An information processing device comprising:

• a memory;
• and a processor that accesses the memory,
• the information processing device being configured to communicate with a reservoir which is supplied with input by the information processor and transitions from a past state to a current state in response to the input,
• the reservoir includes
• an input layer to which the input is applied,
• a first output layer configured to output a plurality of first readout values based on the input,
• a second output layer configured to output a plurality of second readout values based on the input,
• a first partial reservoir including the input layer and the first output layer, and
• a second partial reservoir including the input layer and the second output layer and in which the size between the input layer and the second output layer is larger than the size between the input layer and the first output layer of the first partial reservoir, and
• the processor executes processing for training the reservoir,
• the training processing including:
• first processing
• to update a plurality of first output weights to reduce a difference between
• a first output that is a first product sum value of a plurality of third readout values and the plurality of first output weights, the plurality of third readout values being output by the first output layer based on a first input, and
• teaching data corresponding to the first input, and
• to calculate a differential teaching data that is a difference between a third product sum value of the plurality of third readout values and the updated plurality of first output weights, and
• the teaching data corresponding to the first input;
• and
• second processing to update a plurality of second output weights to reduce a difference between
• a second output that is a second product sum value of a plurality of fourth readout values and the plurality of second output weights, the plurality of fourth readout values being output by the second output layer based on the first input, and
• the differential teaching data,
• the first processing and the second processing are repeated for a plurality of training data respectively including a first input and a teaching data.

A non-transitory computer readable memory medium that stores therein an information processing program causing a processor of an information processing device to execute an information processing,

• the information processing comprising:
• a processing for training a reservoir,
• the information processing device includes a memory, and a processor that accesses the memory, the information processing device being configured to communicate with the reservoir which is supplied with input by the information processor and transitions from a past state to a current state in response to the input, and
• the reservoir includes
• an input layer to which the input is applied,
• a first output layer configured to output a plurality of first readout values based on the input,
• a second output layer configured to output a plurality of second readout values based on the input,
• a first partial reservoir including the input layer and the first output layer, and
• a second partial reservoir including the input layer and the second output layer and in which the size between the input layer and the second output layer is larger than the size between the input layer and the first output layer of the first partial reservoir, and
• the processor executes processing for training the reservoir,
• the training processing including:
• first processing
• to update a plurality of first output weights to reduce a difference between
• a first output that is a first product sum value of a plurality of third readout values and the plurality of first output weights, the plurality of third readout values being output by the first output layer based on a first input, and
• teaching data corresponding to the first input, and
• to calculate a differential teaching data that is a difference between
• a third product sum value of the plurality of third readout values and the updated plurality of first output weights, and
• the teaching data corresponding to the first input;
• and
• second processing to update a plurality of second output weights to reduce a difference between
• a second output that is a second product sum value of a plurality of fourth readout values and the plurality of second output weights, the plurality of fourth readout values being output by the second output layer based on the first input, and
• the differential teaching data,
• the first processing and the second processing are repeated for a plurality of training data respectively including a first input and a teaching data.