Key-value memory network for predicting time-series metrics of target entities

A method comprising:

• receiving, by a processing device for a target entity input data comprising a concatenation of static feature data of the target entity, time-specific feature data, and time-series feature data for the target entity;
• generating, by the processing device, predicted time-series data for a target metric of the target entity by applying a key-value memory network to the input data, the key-value memory network comprising:
• a key matrix with key vectors that are learned from training static feature data and training time-series feature data;
• a value matrix with value vectors representing time-series trends;
• an input layer configured to receive the input data for the target entity;
• an entity-embedding layer configured to generate an input vector from the input data;
• a key-addressing layer configured to generate a weight vector indicating similarities between the key vectors and the input vector;
• a value-reading layer configured to compute a context vector from the weight vector and the value vectors;
• and
• an output layer configured to generate the predicted time-series data for the target metric of the target entity by at least applying a continuous activation function to the context vector and the input vector.

A non-transitory computer readable medium storing computer readable program instructions, that when executed by one or more processing devices, cause the one or more processing devices to perform operations comprising:

• receiving, for a target entity input data comprising a concatenation of static feature data of the target entity, time-specific feature data, and time-series feature data for the target entity;
• generating predicted time-series data for a target metric of the target entity by applying a key-value memory network to the input data, the key-value memory network comprising:
• a key matrix with key vectors that are learned from training static feature data and training time-series feature data;
• a value matrix with value vectors representing time-series trends;
• an input layer configured to receive the input data for the target entity;
• an entity-embedding layer configured to generate an input vector from the input data;
• a key-addressing layer configured to generate a weight vector indicating similarities between the key vectors and the input vector;
• a value-reading layer configured to compute a context vector from the weight vector and the value vectors;
• and
• an output layer configured to generate the predicted time-series data for the target metric of the target entity by at least applying a continuous activation function to the context vector and the input vector.