Systems and methods for enhancing human-in-the-loop labeling process using tabural embedding and label propagation

A non-transitory computer readable medium containing executable instructions, wherein, when executed by a processor, the instructions cause the processor to perform procedures comprising:

• transforming an input tabular dataset comprising a plurality of unlabeled data points and one or more labeled data points, into a plurality of embedded data points, wherein each data sample in the input tabular dataset comprises one or more features, wherein the transforming operation comprises:
• tokenizing the one or more features for each of the plurality of tabular data samples to generate a plurality of feature vectors representing the plurality of tabular data samples in an embedded data space;
• attaching a classification (CLS) token to a beginning of each feature vector, to generate a plurality of input sequences representing each embedded data sample;
• normalizing each of the plurality of input sequences across all features within the input sequence;
• processing each of a plurality of normalized input sequences through one or more transformation layers to generate an optimized embedding space comprising a plurality of multi-dimensional embeddings;
• identifying one or more first embedded data points comprising a target label in the embedded data space, wherein the one or more first embedded data points correspond to the one or more labeled data points in the input tabular dataset;
• and
• generating labels for each of the plurality of unlabeled embedded data points based on a degree of proximity to one or more nearest labeled data point starting from the first embedded data points in the optimized embedding space.

A computer hardware arrangement, comprising:

• a processor;
• and
• a memory containing executable instructions,
• wherein, when executing the instructions, the processor:
• transforms an input tabular dataset comprising a plurality of unlabeled data points and one or more labeled data points, into a plurality of embedded data points, wherein each data sample in the input tabular dataset comprises one or more features, wherein the transforming operation comprises:
• tokenizing the one or more features for each of the plurality of tabular data samples to generate a plurality of feature vectors representing the plurality of tabular data samples in an embedded data space,
• attaching a classification (CLS) token to a beginning of each feature vector, to generate a plurality of input sequences representing each embedded data sample,
• normalizing each of the plurality of input sequences across all features within the input sequence, and
• processing each of a plurality of normalized input sequences through one or more transformation layers to generate an optimized embedding space comprising a plurality of multi-dimensional embeddings;
• identifies one or more first embedded data points comprising a target label in the embedded data space, wherein the one or more first embedded data points correspond to the one or more labeled data points in the input tabular dataset;
• and
• generates labels for each of the plurality of unlabeled embedded data points, based on a degree of proximity to one or more nearest labeled data point starting from the first embedded data points, in the optimized embedding space.