Systems and methods for sample efficient training of machine learning models

A computer program product comprising a non-transitory, computer-readable medium storing thereon a set of computer-executable instructions, the set of computer-executable instructions comprising instructions for:

• receiving an initial set of labeled examples and a set of unlabeled examples, wherein the initial set of labeled examples are labeled according to an abstract category comprising a plurality of semantic concepts;
• mapping an input dataset comprising the initial set of labeled examples and a subset of the unlabeled examples to a reduced-dimension space, the reduced-dimension space comprising labeled data points representing the initial set of labeled examples and unlabeled data points representing the subset of the unlabeled examples;
• identifying a cluster of data points in the reduced-dimension space corresponding to a first semantic concept of the plurality of semantic concepts;
• identifying a target example from the subset of the unlabeled examples based on a corresponding unlabeled data point being located proximate to the cluster of data points corresponding to the first semantic concept;
• assigning a temporary label to the target example indicating a positive match to the first semantic concept;
• providing the temporary label for verification by a user;
• assigning a training label to the target example based on a result of the verification by the user;
• and
• training a machine learning model to classify data according to the plurality of semantic concepts using a final set of labeled examples including the target example.

A system for updating a machine learning model comprising:

• a processor;
• and
• a memory storing instructions that, when executed by the processor, cause the system to:
• receive an initial set of labeled examples and a set of unlabeled examples, wherein the initial set of labeled examples are labeled according to an abstract category comprising a plurality of semantic concepts;
• map an input dataset comprising the initial set of labeled examples and a subset of the unlabeled examples to a reduced-dimension space, the reduced-dimension space comprising labeled data points representing the initial set of labeled examples and unlabeled data points representing the subset of the unlabeled examples;
• identify a cluster of data points in the reduced-dimension space corresponding to a first semantic concept of the plurality of semantic concepts;
• identify a target example from the subset of the unlabeled examples based on a corresponding unlabeled data point being located proximate to the cluster of data points corresponding to the first semantic concept;
• assign a temporary label to the target example indicating a positive match to the first semantic concept;
• provide the temporary label for verification by a user;
• assign a training label to the target example based on a result of the verification by the user;
• and
• train a machine learning model to classify data according to the plurality of semantic concepts using a final set of labeled examples including the target example.