Fujitsu Corporation

Ripon K Saha

Mukul Ranjan Prasad

Hiroaki Yoshida

Indradeep Ghosh

Xiang Gao

Seemanta Saha

Chenguang Zhu

Wenyu Wang

According to one or more embodiments, operations may include normalizing machine learning (ML) pipelines of existing ML projects stored in a corpus of existing ML projects. The operations may also include extracting functional blocks from the normalized ML pipelines. In addition, the operations may include assigning a label to each of the functional blocks in the normalized ML pipelines. Moreover, the operations may include indexing each of the ML pipelines in the corpus based on the labels assigned to the functional blocks. In addition, the operations may include utilizing the labels assigned to the functional blocks in the corpus to generate a new pipeline to perform a new ML task on a new dataset of a new ML project.

Automatically labeling functional blocks in pipelines of existing machine learning projects in a corpus adaptable for use in new machine learning projects

According to one or more embodiments, operations may include storing existing machine learning (ML) projects in a corpus. The operations may also include generating a search query for a new ML project based on a new dataset and a new ML task for the new ML project. In addition, the operations may include searching through the existing ML projects stored in the corpus, based on the search query, for a set of existing ML projects. Moreover, the operations may include merging the ML pipelines of the set of existing ML projects to generate a new ML pipeline for the new ML project. In addition, the operations may include adapting functional blocks of the new ML pipeline for the new ML project to enable the new ML pipeline to be executed to perform the new ML task on the new dataset.

Automatically generating a pipeline of a new machine learning project from pipelines of existing machine learning projects stored in a corpus

According to an aspect of an embodiment, operations may include selecting, from a training dataset, a first data point as a seed data point. The operations may further include generating a population of data points by application of a genetic model on the seed data point. The population of data points may include the seed data point and a plurality of transformed data points of the seed data point. The operations may further include determining a best-fit data point in the generated population of data points based on application of a fitness function on the generated population of data points. The operations may further include executing a training operation on the DNN based on the determined best-fit data point. The operations may further include obtaining a trained DNN for the first data point based on the training operation on the DNN based on the determined best-fit data point.

Data augmentation in training deep neural network (DNN) based on genetic model

According to an aspect of an embodiment, operations may include receiving a first data point associated with a real-time application and predicting a first class for the received first data point, by a Deep Neural Network (DNN) pre-trained for a classification task of the real-time application. The operations may further include extracting, from the DNN, a first set of features and a corresponding first set of weights, for the predicted first class. The extracted first set of features may be associated with a convolution layer of the DNN. The operations may further include determining, by a pre-trained classifier associated with the predicted first class, a confidence score for the predicted first class based on the extracted first set of features and the corresponding first set of weights. The operations may further include generating output information to indicate correctness of the predicted first class based on the determined confidence score.

Validation of deep neural network (DNN) prediction based on pre-trained classifier

According to one or more embodiments, operations may include gathering a set of machine learning (ML) projects from one or more repositories of ML projects based on a filtering criteria. The operations may also include ensuring executability of ML pipelines in the set of ML projects. In addition, the operations may include identifying irrelevant portions of the ML pipelines in the set of ML projects. Moreover, the operations may include generating quality features for the set of ML projects. In addition, the operations may include generating diversity features for the set of ML projects. Moreover, the operations may include selecting a subset of ML projects from the set of ML projects based on the quality features and the diversity features. In addition, the operations may include storing the subset of ML projects in a corpus of ML projects that may be adapted for use in new ML projects.

Automatically curating existing machine learning projects into a corpus adaptable for use in new machine learning projects

Operations include obtaining a machine learning (ML) pipeline skeleton that indicates a set of first functional blocks to use to process a new dataset of a new ML project. The operations also include obtaining a relationship mapping that maps dataset features to respective functional blocks, the relationship mapping indicating correspondences between dataset features of existing datasets of existing ML projects and usage of second functional blocks of existing ML pipelines of the existing ML projects. The operations also include mapping the first functional blocks to respective portions of the new dataset based on the relationship mapping. In addition, the operations include instantiating the pipeline skeleton with respective code snippets that each correspond to a respective first functional block of the set of first functional blocks, the respective code snippets each including one or more respective code elements that are based on the mapping of the first functional blocks.

Machine learning pipeline skeleton instantiation

A method of automated candidate repair patch generation may include synthesizing repair expressions based on project code. Concise repair expressions may be mined from a code database. A set of repair expressions may be generated based on the synthesized repair expressions and the mined repair expressions. The set of repair expressions may include further repair expressions further synthesized from the synthesized repair expressions and the mined repair expressions. Candidate patches may be generated based on a repair schema and the set of repair expressions.

Automated candidate repair patch generation

A method may include obtaining multiple lines of programming code of a program, and obtaining multiple test cases for testing the program, where each of the test cases includes an assertion upon which a result of a respective test case is based. The method may also include executing the program for each of the test cases, and identifying affected lines of programming code that influence the assertions. The method may additionally include calculating a risk score for at least one of the lines of programming code based on the affected lines of programming code and the assertion, the risk score indicative of a likelihood that the at least one of the lines of programming code includes a fault.

Identification of software program fault locations

According to an aspect of an embodiment, a method may include executing multiple tests with respect to code under test of a software program to perform multiple test executions. The method may further include identifying one or more passing tests and one or more failing tests of the test executions. In addition, the method may include determining an aggregated score for each statement based on two or more of: the passing tests and the failing tests; a semantic similarity between one or more statement tokens included in the respective statement and one or more report tokens included in an error report; and an amount of time that has passed from when the respective statement received a change. Moreover, the method may include identifying a particular statement of the plurality of statements as a fault location in the code under test based on the aggregated scores of the plurality of statements.

Software program fault localization

Operations may include obtaining a first code snippet associated with a fault location of a fault of the software program. The operations may further include obtaining a second code snippet. In addition, the operations may include determining element similarity between first elements of the first code snippet and second elements of the second code snippet. Further, the operations may include generating, based on the determined element similarity, an element map that maps the first elements of the first code statement to the second elements of the second code statement. The operations may further include obtaining an abstract program modification as a repair candidate of the fault. In addition, the operations may include generating a first repair based on the abstract program modification and the element map. Moreover, the operations may include generating a second repair based on the abstract program modification and the element map.

Automated software program repair of similar code snippets

The disclosed method may include accessing features including feature information of one or more candidate target projects and of a subject project, in which the candidate target projects and the subject project are software programs. The method may include determining a similarity score between the feature information of each of the candidate target projects and the feature information of the subject project, in which a similarity score is determined for each feature of each of the candidate target projects. The method may include aggregating the similarity scores of the feature information of each feature in the candidate target projects to create an aggregate similarity score for each of the candidate target projects and generate a set of similar target projects. The method may include modifying the subject project by implementing recommended code, based on the similar target projects, in the subject project to repair a defect.

Automatic identification of relevant software projects for cross project learning

A method may include obtaining a plurality of synthesized method invocations using a plurality of objects and a plurality of methods of a software program. The method may also include determining a prioritization of the plurality of synthesized method invocations based on one or more of: relationships between one or more characteristics of each of the plurality of synthesized method invocations and a fault location; and relationships between the one or more characteristics and an error report that corresponds to the fault location. The method may also include selecting a synthesized method invocation from the plurality of synthesized method invocations for repair operations with respect to the fault location based on a corresponding prioritization of the selected synthesized method invocation. In addition, the method may include performing repair operations with respect to the fault location and the selected synthesized method invocation.

Method invocation synthesis for software program repair

In some examples, a method to interactively repair a software program using one or more automatically generated tests with human-provided test oracles may include identifying a fault location in a software program, generating a potential repair at the fault location based on a repair candidate, automatically generating a first test to test the potential repair, and generating a first query for a first test oracle based on the first test. The method may also include obtaining a response to the first query from a human, generating a first human-provided test oracle based on the first query and the obtained response to the first query, augmenting a test suite to include the first automatically generated test with the first human-provided test oracle, and testing the potential repair using the augmented test suite including the first automatically generated test with the first human-provided test oracle.

Interactive software program repair

According to an aspect of an embodiment, a method may include identifying a fault at a fault location in a software program using a test suite. The method may also include determining multiple textual similarity scores by determining a textual similarity score with respect to each of multiple repair candidates for the fault. In addition, the method may include sorting the repair candidates based on the textual similarity scores. The method may also include selecting a particular repair candidate from the repair candidates based on the sorting. Moreover, the method may include implementing the particular repair candidate at the fault location based on the selection of the particular repair candidate.

Growing community of inventors

Santa Clara, CA, United States of America

Ripon K Saha