Systems and methods for transfer learning of neural networks

A computer-implemented method for training a specialized neural network, comprising:

• receiving a first training set;
• determining that the first training set is most similar to a first base cluster in a plurality of base clusters that group data objects included in a second training set, wherein:
• the second training set has a higher number of types of data objects than a number of types of data objects in the first base cluster;
• and
• determining that the first base cluster is most similar to the first training set includes:
• for each cluster of the plurality of base clusters, computing a distance value representing a degree of similarity between values of one or more attributes in data objects of the cluster and values of the one or more attributes in data objects of the first training set by calculating a distance between a mean value of the one or more attributes in data objects of the cluster and a mean value of the one or more attributes in data objects of the first training set;
• identifying a generalized base model trained using the first base cluster, wherein the generalized base model trained using the first base cluster is identified from a plurality of trained generalized base models that were trained using different base clusters;
• and
• training the identified generalized base model using the first training set to obtain the specialized neural network.

A computer system for training a specialized neural network, comprising:

• a memory storing instructions;
• and
• one or more processors configured to execute the instructions to perform operations including:
• receiving a first training set;
• determining that the first training set is most similar to a first base cluster in a plurality of base clusters that group data objects included in a second training set, wherein:
• determining that the first base cluster is most similar to the first training set includes, for each cluster of the plurality of base clusters, computing a distance value representing a degree of similarity between values of one or more attributes in data objects of the cluster and values of the one or more attributes in data objects of the first training set;
• and
• computing the distance value representing the degree of similarity includes calculating a distance between a mean value of the one or more attributes in data objects of the cluster and a mean value of the one or more attributes in data objects of the first training set;
• identifying a generalized base model trained using the first base cluster, wherein the generalized base model trained using the first base cluster is identified from a plurality of trained generalized base models that were trained using different base clusters;
• and
• training the identified generalized base model using the first training set to obtain the specialized neural network.