System and method of training machine-learning-based model

1 . A method of training a machine-learning (ML)-based model by at least one processor, the method comprising:

receiving, by the processor, an initial dataset comprising a plurality of digitally stored annotated data samples each associated with at least one class label;

automatically dividing, by a dataset-analysis module executed by the processor, the initial dataset into a training portion and an inference portion according to class-balance parameters;

training, by a first-level ML-based model implemented on the processor, using the training portion, to perform a first-level classification task;

inferring, by the trained first-level ML-based model, on the inference portion to generate quantitative contribution metrics for each data sample, each metric representing the expected contribution of that sample to accuracy of a second-level ML model;

executing, by an omission module, an automated computation that selects and omits, from the inference portion, a subset of data samples whose contribution metrics satisfy a predefined omission condition comprising (i) association with a majority class and (ii) confidence value exceeding a threshold, thereby generating a target dataset;

iteratively repeating, by the processor, the training, inferring, and omitting operations across multiple iterations with dynamically updated omission percentages until a convergence criterion on dataset balance or model accuracy is met; and

training, by the processor, a second level ML-based model on the resulting target dataset to perform a second-level predictive task with improved classification accuracy relative to training on the unbalanced dataset, wherein the iterative automated omission and retraining process implemented by the modules improves training-dataset quality and thereby enhances reliability and computation performance of the ML-based model.

2 . The method of claim 1 , wherein the digitally stored annotated data samples received by the processor further comprise at least one class label selected based on the amount of data samples associated with each class in the initial dataset.

3 . The method of claim 2 , wherein training the first-level ML-based model comprises:

dividing the initial dataset into a training portion of data samples and an inference portion of data samples;

training the first-level ML-based model to perform the first-level classification task, based on the training portion;

inferring the first-level ML-based model on the inference portion to perform the first-level task, based on the training of the at least one ML-based first-level model.

4 . The method of claim 3 , wherein the division of the initial dataset into the training portion of data samples and the inference portion of data samples is performed by having a random ratio of the data samples of each class of a plurality of classes in the training portion or in the inference portion.

5 . The method of claim 3 , wherein

the first-level classification task comprises classification of the data samples of the initial dataset according to a plurality of classes;

the quantitative contribution metrics comprises a confidence value representing, in result of the inferring, a pertinence of one or more data samples of the inference portion with their respective associated classes; and

wherein the method further comprises

selecting a subset of data samples from the inference portion based on a set of omission conditions, said omission conditions comprising (i) an association of the selected data samples with the selected at least one class in result of inferring, and (ii) a selection of the data samples based on the computated confidence value.

6 . The method of claim 5 , wherein the selection of the data samples based on the computated confidence value comprises the selection of the data samples having (a) a highest-scoring confidence value, and (b) a confidence value that surpasses a predefined threshold.

7 . The method of claim 5 , wherein

training at least one first-level ML-based model comprises training a plurality of first-level ML-based models; and

the selection of the data samples based on the computated confidence value comprises a selection of the data samples based on a function of the computated confidence values of specific data samples, in the plurality of first-level models.

8 . The method of claim 3 , wherein

the first-level task comprises clustering the selected at least one class label and forming, in result of clustering, a set of clusters of the data samples which are associated with the selected at least one class label;

the at least one characteristic comprises a distance of the one or more data samples from a centroid of a cluster of the set of clusters to which the one or more data samples pertain, in result of the inferring; and

wherein the method further comprises

selecting a subset of data samples from the inference portion based on a set of omission conditions, said omission conditions comprising selection of the data samples based on the distance.

9 . The method of claim 8 , wherein the selection of the data samples based on the distance comprises selection of the data samples having (a) a shortest distance to the centroid of the cluster to which the one or more data samples pertain, (b) a distance that is lower than predefined threshold.

10 . The method of claim 8 , wherein the selection of the data samples based on the distance comprises,

for at least one cluster, determination of a range of distances corresponding to a densest distribution of the data samples in the cluster; and

selection of at least one data sample having the distance from the determined range.

11 . The method of claim 3 , wherein the second-level predictive task comprises classification of data samples of an incoming dataset according to a plurality of classes.

12 . The method of claim 3 , wherein the method further comprises

for the selected at least one class, calculating an omission percentage, representing percentage of data samples that are to be omitted from the initial dataset; and

selecting a subset of data samples from the inference portion based on a set of omission conditions, comprising selection of the data samples up to the calculated omission percentage.

13 . A system for training an ML-based model, the system comprising: a non-transitory memory device, wherein modules of instruction code are stored, and at least one processor associated with the memory device, and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor is configured to:

receive, by the processor, an initial dataset comprising a plurality of digitally stored annotated data samples each associated with at least one class label;

automatically divide, by a dataset-analysis module executed by the processor, the initial dataset into a training portion and an inference portion according to class-balance parameters;

train, by a first-level ML-based model implemented on the processor, using the training portion, to perform a first-level classification task;

infer, by the trained first-level ML-based model, on the inference portion to generate quantitate contribution metrics for each data sample, each metric representing the expected contribution of that sample to accuracy of a second-level ML model;

execute, by an omission module, an automated computation that selects and omits, from the inference portion, a subset of data samples whose contribution metrics satisfy a predefined omission condition comprising (i) association with a majority class and (ii) confidence value exceeding a threshold, thereby generating a target dataset;

iteratively repeat, by the processor, the training, inferring, and omitting operations across multiple iterations with dynamically updated omission percentages until a convergence criterion on dataset balance or model accuracy is met; and

train, by the processor, a second level ML-based model on the resulting target dataset to perform a second-level predictive task with improved classification accuracy relative to training on the unbalanced dataset, wherein the iterative automated omission and retraining process implemented by the modules improves training-dataset quality and thereby enhances reliability and computation performance of the ML-based model.

14 . The system of claim 13 , wherein the digitally stored annotated data samples received by the processor further comprise at least one class label selected, based on the amount of data samples associated with each class in the initial dataset.

15 . The system of claim 14 , wherein the at least one processor is configured to train the first-level ML-based model by:

dividing the initial dataset into a training portion of data samples and an inference portion of data samples;

training the first-level ML-based model to perform the first-level classification task, based on the training portion;

inferring the first-level ML-based model on the inference portion to perform the first-level task, based on the training of the at least one ML-based first-level model.

16 . The system of claim 15 , wherein

the first-level classification task comprises classification of the data samples of the initial dataset according to a plurality of classes;

the quantitative contribution metrics comprises a confidence value representing, in result of the inferring, a pertinence of one or more data samples of the inference portion with their respective associated classes; and

wherein the at least one processor is configured to

select a subset of data samples from the inference portion based on a set of omission conditions, said omission conditions comprising (i) an association of the selected data samples with the selected at least one class in result of inferring, and (ii) a selection of the data samples based on the computated confidence value.

17 . The system of claim 16 , wherein the at least one processor is configured to select the data samples based on the computated confidence value as having (a) a highest-scoring confidence value, and (b) a confidence value that surpasses a predefined threshold.

18 . The system of claim 16 , wherein the at least one first-level ML-based model comprises a plurality of first-level ML-based models; and the selection of the data samples based on the computated confidence value comprises a selection of the data samples based on a function of the computated confidence values of specific data samples, in the plurality of first-level ML-based models.

19 . The system of claim 15 , wherein

the first-level task comprises clustering the selected at least one class label and forming, in result of clustering, a set of clusters of the data samples which are associated with the selected at least one class label;

the at least one characteristic comprises a distance of the one or more data samples from a centroid of a cluster of the set of clusters to which the one or more data samples pertain, in result of the inferring; and

wherein the at least one processor is further configured to:

select a subset of data samples from the inference portion based on a set of omission conditions, said omission conditions comprising selection of the data samples based on the distance.