Automated repair of bugs and security vulnerabilities in software

Patent metadata

Field	Value
Publication number	US-10866877-B2
Application number	US-201816189353-A
Country	US
Kind code	B2
Filing date	Nov 13, 2018
Priority date	Nov 13, 2017
Publication date	Dec 15, 2020
Grant date	Dec 15, 2020

Abstract

A software instruction code repair system comprising an instruction code example pool. The example pool comprises a set of good instruction code examples and a set of bad instruction code examples. The software instruction code repair system further comprises a sequence-to-sequence (seq2seq) network that is configured to generate a corrected instruction code example, based on one example of the set of bad instruction code examples. The software instruction code repair system further comprises a discriminator configured to randomly select one of the corrected instruction code example and one instruction code example of the set of good instruction code examples to produce a selected instruction code example. The discriminator is further configured to make a determination that the selected instruction code example is most likely taken either the instruction code example pool or the seq2seq network.

First claim

What is claimed is:

1. A sequential data repair system, comprising:
an example pool of sequential data examples stored on a data storage device, the example pool of sequential data examples comprising a set of good examples and a set of bad examples;
a computer-based sequence-to-sequence (seq2seq) network that comprises a deep learning neural network model, the seq2seq configured to communicate with the data storage device to receive an example of the set of bad examples and to generate a corrected example based on the example of the set of bad examples;
a computer-based discriminator configured to communicate with the data storage device and the seq2seq network, and to:
(a) randomly select either the corrected example or an example of the set of good examples, to produce a selected example, and
(b) make a determination that the selected example was most likely selected from the example pool or the seq2seq network;
(c) generate a training error signal based on the determination, the seq2seq network uses the training error signal to train the deep learning neural network model; and

a user interface configured to present the corrected example to a user.

2. The sequential data repair system of claim 1 , wherein the discriminator and seq2seq network are configured as a generative adversarial network (GAN).

3. The sequential data repair system of claim 1 , wherein the seq2seq network comprises an encoder configured to produce an intermediate representation of an input data sequence, and a decoder configured to produce and output data sequence based on the intermediate representation of the input data sequence.

4. The sequential data repair system of claim 1 , wherein each good example in the set of good examples represents a sequence of data elements having a quality that equals or exceeds a threshold quality level of sequential data associated with the sequential data repair system.

5. The sequential data repair system of claim 1 , wherein each bad example in the set of bad examples represents a sequence of data elements having a quality that is less than or equal to a minimum threshold quality level of sequential data associated with the sequential data repair system.

6. The sequential data repair system of claim 1 , wherein each bad example in the set of bad examples is labeled to indicate a bad example, and each good example in the set of good examples is labeled to indicate a good example.

7. A software instruction code repair system, comprising:
a pool of instruction code examples stored on a data storage device, the pool of instruction code examples comprising a set of good instruction code examples and a set of bad instruction code examples;
a computer-based sequence-to-sequence (seq2seq) network that comprises a deep learning neural network model, the seq2seq configured to communicate with the data storage device to receive an example of the set of bad examples and to generate a corrected instruction code example based on one example of the set of bad instruction code examples;
a computer-based discriminator configured to communicate with the data storage device and the seq2seq network, and to:
(a) randomly select either the corrected instruction code example or an instruction code example of the set of good instruction code examples, to produce a selected instruction code example, and
(b) make a determination that the selected instruction code example was most likely selected from either the instruction code example pool or the seq2seq network;
(c) generate a training error signal based on the determination, the seq2seq network uses the training error signal to train the deep learning neural network model; and

a user interface configured to present the corrected instruction code example to a user.

8. The software instruction code repair system of claim 7 , wherein the discriminator and seq2seq network are configured as a generative adversarial network (GAN).

9. The software instruction code repair system of claim 7 , wherein the seq2seq network comprises an encoder configured to produce an intermediate representation of an input software instruction code, and a decoder configured to produce and output code sequence based on the intermediate representation of the input software instruction code.

10. The software instruction code repair system of claim 7 , wherein each good example in the set of good examples represents a software instruction code example having a quality that equals or exceeds a threshold quality level of software instruction code associated with the software instruction code repair system.

11. The software instruction code repair system of claim 7 , wherein each bad example in the set of bad examples represents a software instruction code example having a quality that is less than or equal to a minimum threshold quality level of software instruction code associated with the software instruction code repair system.

12. The software instruction code repair system of claim 7 , wherein each bad example in the set of bad examples is labeled to indicate a bad software instruction code example, and each good example in the set of good examples is labeled to indicate a good software instruction code example.

13. A method of repairing a sequence of data elements, comprising:
storing, on a data storage device, a set of good instruction code examples and a set of bad instruction code examples;
generating, using a computer-based sequence-to-sequence (seq2seq) network that comprises a deep learning neural network model, the seq2seq configured to communicate with the data storage device, a corrected instruction code example based on one example of the set of bad instruction code examples;
randomly selecting, using a computer-based discriminator, either the corrected instruction code example or an instruction code example of the set of good instruction code examples, to produce a selected instruction code example,
making a determination, using the computer-based discriminator, that the selected instruction code example was most likely selected from either the instruction code example pool or the seq2seq network;
generating a training error signal based on the determination, the seq2seq network uses the training error signal to train the deep learning neural network model; and
presenting, using a user interface, the corrected instruction code example.

14. The method of claim 13 , further comprising configuring the discriminator and seq2seq network as a generative adversarial network (GAN).

Assignees

• Charles Stark Draper Laboratory Inc

Inventors

• Harer Jacob
• Lazovich Tomo
• Russell Rebecca
• Ozdemir Onur
• Kim Louis

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/094

  Adversarial learning · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/08

  Learning methods · CPC title