Staged training of neural networks for improved time series prediction performance

A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, the computer-program product including instructions operable to cause a processor to perform operations comprising:

• train, using initial neural network configuration data comprising neural network hyperparameters, a first neural network of a chain of neural networks to generate first neural network configuration data comprising the hyperparameters and first trained parameters learned by the first neural network, wherein:
• the chain is to perform an analytical function to generate a set of output data values from a set of input data values;
• the analytical function comprises the generation of a time series prediction that covers a selected full range of time;
• the chain comprises a set of multiple neural networks that includes at least the first neural network and a last neural network;
• the set of neural networks is ordered to form the chain starting with the first neural network at a head of the chain and ending with the last neural network at a tail of the chain;
• each neural network in the chain comprises external inputs to receive the set of input data values;
• each neural network in the chain comprises outputs at which the neural network outputs a portion of the set of output data values from the set input data values during operation of the chain to perform the analytical function;
• and
• the set of neural networks is interconnected within the chain such that each neural network in the chain, except the first neural network at the head of the chain, receives the outputs of a preceding neural network in the ordering of neural networks within the chain as additional inputs;
• train, using the first neural network configuration data, a next neural network in the ordering of neural networks within the chain to generate a next neural network configuration data comprising the hyperparameters and next trained parameters learned by the next neural network;
• use at least the first neural network configuration data, the next neural network configuration data, and additional data comprising an indication of interconnections among the neural networks within the chain to instantiate the chain of neural networks;
• and
• operate the chain of neural networks to perform the analytical function.

A computer-implemented method comprising:

• training, by a processor, and using initial neural network configuration data comprising neural network hyperparameters, a first neural network of a chain of neural networks to generate first neural network configuration data comprising the hyperparameters and first trained parameters learned by the first neural network, wherein:
• the chain is to perform an analytical function to generate a set of output data values from a set of input data values;
• the analytical function comprises the generation of a time series prediction that covers a selected full range of time;
• the chain comprises a set of multiple neural networks that includes at least the first neural network and a last neural network;
• the set of neural networks is ordered to form the chain starting with the first neural network at a head of the chain and ending with the last neural network at a tail of the chain;
• each neural network in the chain comprises external inputs to receive the set of input data values;
• each neural network in the chain comprises outputs at which the neural network outputs a portion of the set of output data values from the set input data values during operation of the chain to perform the analytical function;
• and
• the set of neural networks is interconnected within the chain such that each neural network in the chain, except the first neural network at the head of the chain, receives the outputs of a preceding neural network in the ordering of neural networks within the chain as additional inputs;
• training, by the processor, and using the first neural network configuration data, a next neural network in the ordering of neural networks within the chain to generate a next neural network configuration data comprising the hyperparameters and next trained parameters learned by the next neural network;
• using at least the first neural network configuration data, the next neural network configuration data, and additional data comprising an indication of interconnections among the neural networks within the chain to instantiate the chain of neural networks;
• and
• operating the chain of neural networks to perform the analytical function.