International Business Machines Corporation

Samsung Electronics Co., Ltd.

Saswati Dana

Dinesh Garg

Dinesh Khandelwal

G P Shrivatsa Bhargav

L Venkata Subramaniam

Jonathan D Dunne

Zachary A Silverstein

Trudy L Hewitt

Taeyong Kim

Sreyash Kenkre

Saneem Ahmed Chemmengath

Kyusang Lee

Pavan Kapanipathi Bangalore

Anirban Bhaduri

Alexander Gray

Salim E Roukos

Mayoore Selvarasa Jaiswal

Varun Giri

Venkata Tadi Siva Kumar

Sukannya Purkayastha

A modular two-stage neural architecture is used in translating a natural language question into a logic form such as a SPARQL Protocol and RDF Query Language (SPARQL) query. In a first stage, a neural machine translation (NMT)-based sequence-to-sequence (Seq2Seq) model translates a question into a sketch of the desired SPARQL query called a SPARQL silhouette. In a second stage a neural graph search module predicts the correct relations in the underlying knowledge graph.

Knowledge graph question answering with neural machine translation

An embodiment includes decomposing a natural language assertion into a natural language question and answer pair that includes an initial question and an initial answer. The embodiment translates the initial question into a structured knowledge graph query and then performs an iterative process comprising iterative querying of a knowledge graph and evaluating of corresponding query responses resulting in respective confidence scores. A first iteration of the iterative process comprises querying of the knowledge graph to retrieve a first predicted answer, then determining whether a degree of similarity between the initial answer and the first predicted answer meets a threshold criterion. If not, the first predicted query is altered and used for querying the knowledge graph in a subsequent iteration of the iterative process. The embodiment also generates an assertion correctness score indicative of a degree of confidence that the assertion is factual using the respective confidence scores.

Automated fact checking using iterative knowledge base querying

Methods, systems, and computer program products for zero-shot entity linking based on symbolic information are provided herein. A computer-implemented method includes obtaining a knowledge graph comprising a set of entities and a training dataset comprising text samples for at least a subset of the entities in the knowledge graph; training a machine learning model to map an entity mention substring of a given sample of text to one corresponding entity in the set of entities, wherein the machine learning model is trained using a multi-task machine learning framework using symbolic information extracted from the knowledge graph; and mapping an entity mention substring of a new sample of text to one of the entities in the set using the trained machine learning model.

Zero-shot entity linking based on symbolic information

Methods, systems, and computer program products for explaining cross domain model predictions are provided herein. A computer-implemented method includes providing a test data point to a domain adaptation model to obtain a prediction, wherein the domain adaptation model is trained on a set of labeled data points and a set of unlabeled data points. The method includes generating a task specific explanation for the prediction that includes one or more data points from among the sets that satisfy a threshold score for influencing the prediction. Additionally, the method includes generating a domain invariant explanation for the prediction. The domain variation explanation is generated by ranking pairs of data points based on a statistical similarity to the test data point, wherein each pair includes a data point from the set of labeled data points and a data point from the set of unlabeled data points.

Explaining cross domain model predictions

A computer-implemented method is provided. The embodiments include evaluating, by one or more processors, a specimen chart relative to a chart erratum model that has features mapped to an optimum state for a first chart type. The method also includes generating a first risk score for a first sample feature of the specimen chart. The first risk score may include a delta from the optimum state. The method also includes refactoring the specimen chart to mitigate the first risk score of the first sample feature.

Chart erratum and refactoring

Provided are a method and apparatus for designing and processing a rule pipeline for in silico prediction of chemical reactions. The method includes designing a rule pipeline from at least one rule for chemical conversion and processing at least one input molecule by using the designed rule pipeline to predict a chemical reaction based on a processing result of the processing.

Methods and apparatus for in silico prediction of chemical reactions

Growing community of inventors

Bengaluru, India

Saswati Dana