[highlight]repairing deep neural networks fix patterns and challenges
Published:
http://web.cs.iastate.edu/~design/papers/ICSE-20a/bugrepair.pdf
problems
A significant SE problem in the software that uses DNNs is the presence of bugs. What are the common bugs in such software?
steps
we conduct a comprehensive study of bug repair patterns for five DNN libraries Caffe, Keras, Tensorflow, Theano, and Torch.
We leverage the dataset of DNN bugs published by Islam et al. [10] that consists of 415 bugs from Stack Overflow and 555 bugs from Github
related work
- Zhang et al. [35] have identified bug types, root causes, and their effects in Tensorflow library for DNN
An empirical study on TensorFlow program bugs. In Proceedings of the 27th ACM SIGSOFT International Symposium on Software Testing and Analysis. ACM, 129–140.
- Islam et al. [10] have studied an even larger set of libraries including Caffe, Keras, Tensorflow, Theano, and Torch to identify bug characteristics.
Md Johirul Islam, Giang Nguyen, Rangeet Pan, and Hridesh Rajan. 2019. A omprehensive Study on Deep Learning Bug Characteristics. In Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE 2019). ACM, New York, NY, USA, 510–520. https://doi.org/10.1145/3338906.3338955