Transfer learning based on cross-domain homophily influences

A computer program product comprising:

• a computer readable storage medium readable by one or more processor and storing instructions for execution by the one or more processor for performing a method comprising:
• generating a plurality of deep transfer learning networks based on a plurality of narrow exemplars and a plurality of broad exemplars, wherein a first subset of the narrow exemplars represents a source domain, a second subset of the narrow exemplars represents a target domain, and the broad exemplars represent both the source domain and the target domain;
• encoding one or more transfer layers to a chromosome for genetic operators, where the one or more transfer layers are to be transferred from a source deep transfer learning network corresponding to the source domain to a target deep transfer learning network corresponding to the target domain, wherein the plurality of the deep transfer learning network include both the source deep transfer learning network and the target deep transfer learning network;
• diversifying concurrently both the source deep transfer learning network and the target deep transfer learning network by use of the genetic operators;
• and
• producing the target deep transfer learning network that integrates a result from the diversifying on the source deep transfer learning network and that passes a predefined fitness threshold condition for the target deep transfer learning network.

A system comprising:

• a memory;
• one or more processor in communication with memory;
• and
• program instructions executable by the one or more processor via the memory to perform a method comprising:
• generating a plurality of deep transfer learning networks based on a plurality of narrow exemplars and a plurality of broad exemplars, wherein a first subset of the narrow exemplars represents a source domain, a second subset of the narrow exemplars represents a target domain, and the broad exemplars represent both the source domain and the target domain;
• encoding one or more transfer layers to a chromosome for genetic operators, where the one or more transfer layers are to be transferred from a source deep transfer learning network corresponding to the source domain to a target deep transfer learning network corresponding to the target domain, wherein the plurality of the deep transfer learning network include both the source deep transfer learning network and the target deep transfer learning network;
• diversifying concurrently both the source deep transfer learning network and the target deep transfer learning network by use of the genetic operators;
• and
• producing the target deep transfer learning network that integrates a result from the diversifying on the source deep transfer learning network and that passes a predefined fitness threshold condition for the target deep transfer learning network, wherein the predefined fitness threshold condition is determined based on the accuracy of the target deep transfer learning network as determined from an average between narrow ground truths specific to the target domain and broad ground truths applicable for both the source domain and the target domain.

A computer implemented method comprising:

• generating, by one or more processor, a plurality of deep transfer learning networks based on a plurality of narrow exemplars and a plurality of broad exemplars, wherein a first subset of the narrow exemplars represents a source domain, a second subset of the narrow exemplars represents a target domain, and the broad exemplars represent both the source domain and the target domain;
• training, by the one or more processor, the source deep transfer learning network with the first subset of the narrow exemplars and the broad exemplars;
• encoding, by the one or more processor, one or more transfer layers to a chromosome for genetic operators, where the one or more transfer layers are to be transferred from a source deep transfer learning network corresponding to the source domain to a target deep transfer learning network corresponding to the target domain, wherein the plurality of the deep transfer learning network include both the source deep transfer learning network and the target deep transfer learning network;
• mutating, by the one or more processor, randomly the transfer layers of the source deep transfer learning network from the encoding;
• and
• ascertaining, by the one or more processor, that a fitness threshold condition for the source deep transfer learning network is satisfied;
• migrating, by the one or more processor, weights of respective transfer layers of two or more source deep transfer learning networks over to the target deep transfer learning network respectively.

A computer implemented method comprising:

• generating, by one or more processor, a plurality of deep transfer learning networks based on a plurality of narrow exemplars and a plurality of broad exemplars, wherein a first subset of the narrow exemplars represents a source domain, a second subset of the narrow exemplars represents a target domain, and the broad exemplars represent both the source domain and the target domain;
• training, by the one or more processor, the target deep transfer learning network with the second subset of the narrow exemplars and the broad exemplars;
• encoding, by the one or more processor, one or more transfer layers to a chromosome for genetic operators, where the one or more transfer layers are to be transferred from a source deep transfer learning network corresponding to the source domain to a target deep transfer learning network corresponding to the target domain, wherein the plurality of the deep transfer learning network include both the source deep transfer learning network and the target deep transfer learning network;
• ascertaining, by the one or more processor, that respective transfer layers of two source deep transfer learning networks have their respective weights migrated over to the target deep transfer learning network for respectively passing a fitness threshold condition for the source deep transfer learning network;
• crossing over, by the one or more processor, the transfer layers of the target deep transfer learning network with the transfer layers of the two source deep transfer learning networks from the ascertaining;
• and
• producing, by the one or more processor, the target deep transfer learning network that integrates a result from the crossing over on the target deep transfer learning network and that passes a predefined fitness threshold condition for the target deep transfer learning network.