| Presentation | 2021-03-06 Research for finding faults in Programs using object detection algorithm by CNN-BI system Kazuhiko Ogawa, Takako Nakatani, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In order to predict the location of program faults, we generated images the source code of a faulty program and trained it with a deep learning algorithm applied to object detection to see if We found program fragments where faults may exist. We found that the program statements that cause of faults have something in common in the appearance of the source code, and that we could find the faults by applying CNN (Convolutional Neural Network), a type of deep learning. For the object detection algorithm, we used YOLO (You Look Only Ones), a CNN-based real-time object detection algorithm. In this paper, we attempted to improve the accuracy of inference and to identify the location of faults, which has been a problem in inferring faults. Our goal is to improve the accuracy by using an object detection algorithm to infer the location of faults, instead of the image recognition method we have used so far. In order for the learning models to infer the faults in the program, they learn the actual faults as source code fragments. After training, we use the learning model to reason about the subject's five programs. In the inference, we used bounding boxes to enclose faults in the source code of the subject's programs, just as an object detection algorithm would detect animals or artifacts. In our experiments, we verify that our method improves the accuracy of inferring faults in programs and points out the fault compared to previous inference methods using image recognition. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | bug inference / convolutional nural network / image of source code / deep learning / object detection |
| Paper # | KBSE2020-45 |
| Date of Issue | 2021-02-26 (KBSE) |
| Conference Information | |
| Committee | KBSE |
|---|---|
| Conference Date | 2021/3/5(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Hiroyuki Nakagawa(Osaka Univ.) |
| Vice Chair | Takuya Saruwatari(NTT Data) |
| Secretary | Takuya Saruwatari(OKI) |
| Assistant | Shinpei Ogata(Shinshu Univ.) / Erina Nakihara(Doshisha Univ,) |
| Paper Information | |
| Registration To | Technical Committee on Knowledge-Based Software Engineering |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Research for finding faults in Programs using object detection algorithm by CNN-BI system |
| Sub Title (in English) | |
| Keyword(1) | bug inference |
| Keyword(2) | convolutional nural network |
| Keyword(3) | image of source code |
| Keyword(4) | deep learning |
| Keyword(5) | object detection |
| 1st Author's Name | Kazuhiko Ogawa |
| 1st Author's Affiliation | The Open University of Japan(OUJ) |
| 2nd Author's Name | Takako Nakatani |
| 2nd Author's Affiliation | The Open University of Japan(OUJ) |
| Date | 2021-03-06 |
| Paper # | KBSE2020-45 |
| Volume (vol) | vol.120 |
| Number (no) | KBSE-423 |
| Page | pp.pp.65-70(KBSE), |
| #Pages | 6 |
| Date of Issue | 2021-02-26 (KBSE) |