17808281. QUERY INTERPRETER TRAINING WITH ADVERSARIAL TABLE PERTURBATIONS simplified abstract (Microsoft Technology Licensing, LLC)
Contents
QUERY INTERPRETER TRAINING WITH ADVERSARIAL TABLE PERTURBATIONS
Organization Name
Microsoft Technology Licensing, LLC
Inventor(s)
QUERY INTERPRETER TRAINING WITH ADVERSARIAL TABLE PERTURBATIONS - A simplified explanation of the abstract
This abstract first appeared for US patent application 17808281 titled 'QUERY INTERPRETER TRAINING WITH ADVERSARIAL TABLE PERTURBATIONS
Simplified Explanation
The patent application describes a method for generating adversarial training examples to train a query interpreter. Here are the key points:
- The method starts by obtaining a target data table for a natural language query.
- The primary entity of the target data table is identified.
- A set of candidate identifiers is generated for a target domain associated with the target data table.
- Each candidate identifier is evaluated using an NLI (Natural Language Inference) model to generate an entailment score.
- A subset of candidate identifiers is selected based on the entailment scores.
- Each candidate identifier from the subset is applied as a table perturbation to the target data table, generating a perturbed data table.
- The perturbed data tables are outputted as adversarial training examples.
Potential applications of this technology:
- Improving the accuracy and performance of query interpreters in natural language processing systems.
- Enhancing the training process of query interpreters by generating diverse and challenging training examples.
Problems solved by this technology:
- Lack of diverse and challenging training examples for query interpreters.
- Difficulty in generating realistic and representative perturbations to data tables.
Benefits of this technology:
- Improved accuracy and performance of query interpreters.
- Enhanced training process leading to better understanding and interpretation of natural language queries.
- Increased robustness of query interpreters to handle a wide range of input variations.
Original Abstract Submitted
A set of adversarial training examples for training a query interpreter are generated by: obtaining a target data table for a natural language query; identifying a primary entity of the target data table; for a target domain of the target data table, generating a set of candidate identifiers that are each semantically associated with an identifier of the target domain; for each candidate identifier, providing a premise-hypothesis pair to an NLI model to generate an entailment score; selecting a first subset of candidate identifiers from among the set of candidate identifiers based on the entailment score generated for each premise-hypothesis pair; for each candidate identifier of the first subset, applying the candidate identifier to an instance of the target data table as a table perturbation to generate a perturbed data table; and outputting each perturbed data table as part of an adversarial training example.
