Methods and systems that use incomplete training data to train machine-learning based systems
AssigneeSymbolcraft LLC
InventorPaton J. Lewis
Life of this patent
Estimating the term…
Abstract
The current document is directed to methods and systems that effectively and efficiently employ incomplete training data to train machine-learning-based systems. Incomplete training data, as one example, may include training data with erroneous or inaccurate input-vector/label pairs. In currently disclosed methods and systems, Incomplete training data is mapped to loss classes based on addition training-data information and specific, different additional-information-dependent loss-generation methods are employed for training data of different loss classes during machine-learning-based-system training so that incomplete training data can be effectively and efficiently used.
Claims20
A machine-learning system comprising:
- one or more computer systems, each including one or more processors and one or more memories;
- machine-learning data structures that include weights that are determined during training of the machine-learning system;
- a machine-learning control program that, when executed on one or more of the one or more processors, receives input data, processes the input data using the machine-learning data structures to produce result data, and generates output data that is stored in one or more of the one or more memories;
- and
- a training control program that, when executed on one or more of the one or more processors, iteratively
- selects a next portion of an augmented training dataset that includes input/label pairs, each input/label pair comprising an input and a label,
- maps the selected next portion of the augmented training dataset to loss classes, and
- for each loss class,
- inputs the input/label pairs of the next portion of the augmented training dataset, mapped to the loss class, to the machine-learning control program,
- applies a loss-class-specific method, to output data generated by the machine-learning control program and to labels of the input/label pairs of the next portion of the augmented training dataset mapped to the loss class, to generate a loss, and
- back-propagates the loss through the machine-learning data structures to adjust the included weights.
A method that trains a machine-learning component of a system that includes one or more computer systems, each including one or more processors and one or more memories, machine-learning data structures that include weights that are determined during training of the machine-learning component, and a machine-learning control program that, when executed on one or more of the one or more processors, receives input data, processes the input data using the machine-learning data structures to produce result data, and generates output data that is stored in one or more of the one or more memories, the method comprising:
- receiving a training dataset comprising input/label pairs, each input/label pair comprising an input and a label;
- receiving additional information for one or more of the input/label pairs;
- augmenting the training dataset by, for each input/label pair of the one or more of the input/label pairs, associating additional information for the input/label pair with the input/label pair;
- and
- iteratively
- selecting a next portion of the augmented training dataset,
- mapping the selected next portion of the augmented training dataset to loss classes, and
- for each loss class,
- inputting input/label pairs of the next portion of the augmented training dataset mapped to the loss class to the machine-learning control program,
- applying a loss-class-specific method, to output data generated by the machine-learning control program and to labels of the input/label pairs of the next portion of the augmented training dataset mapped to the loss class, to generate a loss, and
- back-propagating the loss through the machine-learning data structures to adjust the included weights.
A physical data-storage device encoded with computer instructions that, when executed on one or more processors of one or more computer systems that comprise a system that includes a machine-learning component, machine-learning data structures that include weights that are determined during training of the machine-learning component, and a machine-learning control program that receives input data, processes the input data using the machine-learning data structures to produce result data, and generates output data that is stored in one or more memories, controls the one or more computer systems to:
- receive a training dataset comprising input/label pairs, each input/label pair comprising an input and a label;
- receive additional information for one or more of the input/label pairs;
- augment the training dataset by, for each input/label pair of the one or more of the input/label pairs, associating additional information for the input/label pair with the input/label pair;
- and
- iteratively
- select a next portion of the augmented training dataset,
- mapping the selected next portion of the augmented training dataset into loss classes, and
- for each loss class,
- input input/label pairs of the next portion of the augmented training dataset mapped to the loss class to the machine-learning control program,
- apply a loss-class-specific method, to output data generated by the machine-learning control program and to labels of the input/label pairs of the next portion of the augmented training dataset mapped to the loss class, to generate a loss, and
- back-propagate the loss through the machine-learning data structures to adjust the included weights.
Dependencies are inferred from the claim text.