An Empirical Study of Common Challenges in Developing Deep Learning Applications

Tianyi Zhang, Cuiyun Gao, Lei Ma, Michael Lyu, Miryung Kim

研究成果: Chapter in Book/Report/Conference proceedingConference contribution

16 被引用数 (Scopus)

抄録

Recent advances in deep learning promote the innovation of many intelligent systems and applications such as autonomous driving and image recognition. Despite enormous efforts and investments in this field, a fundamental question remains under-investigated - what challenges do developers commonly face when building deep learning applications? To seek an answer, this paper presents a large-scale empirical study of deep learning questions in a popular Q&A website, Stack Overflow. We manually inspect a sample of 715 questions and identify seven kinds of frequently asked questions. We further build a classification model to quantify the distribution of different kinds of deep learning questions in the entire set of 39,628 deep learning questions. We find that program crashes, model migration, and implementation questions are the top three most frequently asked questions. After carefully examining accepted answers of these questions, we summarize five main root causes that may deserve attention from the research community, including API misuse, incorrect hyperparameter selection, GPU computation, static graph computation, and limited debugging and profiling support. Our results highlight the need for new techniques such as cross-framework differential testing to improve software development productivity and software reliability in deep learning.

本文言語英語
ホスト出版物のタイトルProceedings - 2019 IEEE 30th International Symposium on Software Reliability Engineering, ISSRE 2019
編集者Katinka Wolter, Ina Schieferdecker, Barbara Gallina, Michel Cukier, Roberto Natella, Naghmeh Ivaki, Nuno Laranjeiro
出版社IEEE Computer Society
ページ104-115
ページ数12
ISBN(電子版)9781728149813
DOI
出版ステータス出版済み - 10 2019
イベント30th IEEE International Symposium on Software Reliability Engineering, ISSRE 2019 - Berlin, ドイツ
継続期間: 10 28 201910 31 2019

出版物シリーズ

名前Proceedings - International Symposium on Software Reliability Engineering, ISSRE
2019-October
ISSN(印刷版)1071-9458

会議

会議30th IEEE International Symposium on Software Reliability Engineering, ISSRE 2019
Countryドイツ
CityBerlin
Period10/28/1910/31/19

All Science Journal Classification (ASJC) codes

  • Software
  • Safety, Risk, Reliability and Quality

フィンガープリント 「An Empirical Study of Common Challenges in Developing Deep Learning Applications」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル