Abstract: Methods, systems, and computer-readable storage media for a machine learning (ML) system for matching a query entity to one or more target entities, the ML system that reducing a number of query-target entity pairs from consideration as potential matches during inference.

Abstract: Computer-readable media, methods, and systems are disclosed for feature attribution in a machine learning model. Samples may be generated for a machine learning model based on a normalized probability distribution. The samples may be used to determine a weight for features and feature pairs for the machine learning model. The weights of the features and feature pairs may be used to determine which features are significant for predictions within the machine learning model.

Abstract: Methods, systems, and computer-readable storage media for receiving a set of inference results generated by a ML model, the inference results including a set of query entities and a set of target entities, each query entity having one or more target entities matched thereto by the ML model, processing the set of inference results to generate a set of matched sub-sets of target entities by executing a search over target entities in the set of target entities based on constraints, for each problem in a set of problems, providing the problem as a tuple including an index value representative of a target entity in the set of target entities and a value associated with the query entity, the value including a constraint relative to the query entity, and executing at least one task in response to one or more matched sub-sets in the set of matched sub-sets.

Abstract: Methods, systems, and computer-readable storage media for receiving a set of inference results generated by a ML model, the inference results including a set of query entities and a set of target entities, each query entity having one or more target entities matched thereto by the ML model, processing the set of inference results to generate a set of matched sub-sets of target entities by executing a search over target entities in the set of target entities based on constraints, for each problem in a set of problems, providing the problem as a tuple including an index value representative of a target entity in the set of target entities and a value associated with the query entity, the value including a constraint relative to the query entity, and executing at least one task in response to one or more matched sub-sets in the set of matched sub-sets.

Abstract: Methods, systems, and computer-readable storage media for receiving IRF data sets, the IRF data sets including a set of records including inference results determined by the ML model during production use of the ML model and at least one correction to an inference result, executing incremental training of the ML model to provide an updated ML model by selectively filtering one or more records of the set of records to adjust a negative sample to positive sample proportion of a sub-set of records based on a negative sample to positive sample proportion of initial training of the ML model, for each record in the sub-set of records, determining a weight, and during incremental training, applying the weight of a respective record being in a loss function in determining an accuracy of the ML model based on the respective record, and deploying the updated ML model for production use.

Abstract: Pairwise entity matching systems and methods are disclosed herein. A deep learning model may be used to match entities from separate data tables. Entities may be preprocessed to fuse textual and numeric data early in the neural network architecture. Numeric data may be represented as a vector of a geometrically progressing function. By fusing textual and numeric data, including dates, early in the neural network architecture the neural network may better learn the relationships between the numeric and textual data. Once preprocessed, the paired entities may be scored and matched using a neural network.

Abstract: Pairwise entity matching systems and methods are disclosed herein. A deep learning model may be used to match entities from separate data tables. Entities may be preprocessed to fuse textual and numeric data early in the neural network architecture. Numeric data may be represented as a vector of a geometrically progressing function. By fusing textual and numeric data, including dates, early in the neural network architecture the neural network may better learn the relationships between the numeric and textual data. Once preprocessed, the paired entities may be scored and matched using a neural network.

Abstract: In an example embodiment, a deep neural network may be utilized to determine matches between candidate pairs of entities, as well as confidence scores that reflect how certain the deep neural network is about the corresponding match. The deep neural network is also able to find these matches without requiring domain knowledge that would be required if features for a machine-learned model were handcrafted, which is a drawback of prior art machine-learned models used to match entities in multiple tables. Thus, the deep neural network improves on the functioning of prior art machine learned models designed to perform the same tasks. Specifically, the deep neural network learns the relationships of tabular fields and the patterns that define a match from historical data alone, making this approach generic and applicable independent of the context.

Abstract: Disclosed herein are system, method, and computer program product embodiments for generating document labels using positional embeddings. In an embodiment, a label system may identify tokens, such as words, of a document image. The label system may apply a position vector neural network to the document image to analyze the pixels and determine positional embedding vectors corresponding to the words. The label system may then combine the positional embedding vectors to corresponding word vectors for use as an input to a neural network trained to generate document labels. This combination may embed the positional information with the corresponding word information in a serialized manner for processing by the document label neural network. Using this formatting, the label system may generate document labels in a light-weight and fast manner while still preserving spatial relationships between words.

Abstract: Disclosed herein are system, method, and computer program product embodiments for generating document labels using positional embeddings. In an embodiment, a label system may identify tokens, such as words, of a document image. The label system may apply a position vector neural network to the document image to analyze the pixels and determine positional embedding vectors corresponding to the words. The label system may then combine the positional embedding vectors to corresponding word vectors for use as an input to a neural network trained to generate document labels. This combination may embed the positional information with the corresponding word information in a serialized manner for processing by the document label neural network. Using this formatting, the label system may generate document labels in a light-weight and fast manner while still preserving spatial relationships between words.