Triage of training data for acceleration of large-scale machine learning

A computer program product for training an artificial neural network, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method comprising:

• providing training input from a set of training data to an artificial neural network, the artificial neural network comprising a plurality of output neurons, each output neuron corresponding to a class;
• determining, from the artificial neural network, output values at each of the plurality of output neurons;
• determining, from the output values a classification of the training input by the artificial neural network;
• determining a confidence value of the classification by comparing the output values to an output value that corresponds to a predetermined class of the training input;
• based on the confidence value, determining a probability of inclusion of the training input in subsequent training;
• generating a subset of the set of training data based on the probability;
• and
• training the artificial neural network based on the subset.

A method of enhancing the performance of an artificial neural network by identification of a subset of training images to focus on, wherein

• the artificial neural network comprises a plurality of neurons, and is adapted to process a plurality of inputs to provide one or more output,
• the plurality of neurons is arranged in layers, outputs of one of the layers being connected to inputs of each neuron of a subsequent one of the layers by synapses, each of the synapses having a variable strength, the method comprising:
• serially inputting a plurality of training examples to the artificial neural network, a desired output neuron being predetermined for each of the plurality of training examples;
• determining the output of the artificial neural network;
• computing, for each of the training examples a confidence value, the confidence value indicating the performance of the desired output neuron relative to the other output neurons;
• computing a probability from each of the confidence values, indicating the probability of inclusion in subsequent training;
• selecting a subset of the training examples for subsequent training based on the probabilities;
• updating the synaptic strengths by backpropagation based on (i) the difference between the output of the network and (ii) desired output for the subset of the training examples;
• and
• recomputing confidence values, reselecting a subset of the training examples, and re-updating the synaptic strengths until the network achieves a desired target accuracy.