DeepLocalize: Fault Localization for Deep Neural Networks

By: Mohammad Wardat, Wei Le, and Hridesh Rajan

Download Paper

Abstract

Deep neural networks (DNNs) are becoming an integral part of most software systems. Previous work has shown that DNNs have bugs. Unfortunately, existing debugging techniques don’t support localizing DNN bugs because of the lack of understanding of model behaviors. The entire DNN model appears as a black box. To address these problems, we propose an approach and a tool that automatically determines whether the model is buggy or not, and identifies the root causes for DNN errors. Our key insight is that historic trends in values propagated between layers can be analyzed to identify faults, and also localize faults. To that end, we first enable dynamic analysis of deep learning applications: by converting it into an imperative representation and alternatively using a callback mechanism. Both mechanisms allows us to insert probes that enable dynamic analysis over the traces produced by the DNN while it is being trained on the training data. We then conduct dynamic analysis over the traces to identify the faulty layer or hyperparameter that causes the error. We propose an algorithm for identifying root causes by capturing any numerical error and monitoring the model during training and finding the relevance of every layer/parameter on the DNN outcome. We have collected a benchmark containing 40 buggy models and patches that contain real errors in deep learning applications from Stack Overflow and GitHub. Our benchmark can be used to evaluate automated debugging tools and repair techniques. We have evaluated our approach using this DNN bug-and-patch benchmark, and the results showed that our approach is much more effective than the existing debugging approach used in the state-of-the-practice Keras library. For 34/40 cases, our approach was able to detect faults whereas the best debugging approach provided by Keras detected 32/40 faults. Our approach was able to localize 21/40 bugs whereas Keras did not localize any faults.

ACM Reference

Wardat, M. et al. 2021. DeepLocalize: Fault Localization for Deep Neural Networks. ICSE’21: The 43nd International Conference on Software Engineering (May 2021).

BibTeX Reference

@inproceedings{wardat21deeplocalize,
  author = {Mohammad Wardat and Wei Le and Hridesh Rajan},
  title = {DeepLocalize: Fault Localization for Deep Neural Networks},
  booktitle = {ICSE'21: The 43nd International Conference on Software Engineering},
  location = {Virtual Conference},
  month = {May 23-May 29, 2021},
  year = {2021},
  entrysubtype = {conference},
  abstract = {
    Deep neural networks (DNNs) are becoming an integral part of most
    software systems. Previous work has shown that DNNs have bugs.
    Unfortunately, existing debugging techniques don’t support localizing
    DNN bugs because of the lack of understanding of model behaviors.
    The entire DNN model appears as a black box. To address these problems,
    we propose an approach and a tool that automatically determines whether
    the model is buggy or not, and identifies the root causes for DNN errors.
    Our key insight is that historic trends in values propagated between
    layers can be analyzed to identify faults, and also localize faults.
    To that end, we first enable dynamic analysis of deep learning
    applications: by converting it into an imperative representation and
    alternatively using a callback mechanism. Both mechanisms allows us
    to insert probes that enable dynamic analysis over the traces produced
    by the DNN while it is being trained on the training data.
    We then conduct dynamic analysis over the traces to identify the faulty
    layer or hyperparameter that causes the error. We propose an algorithm
    for identifying root causes by capturing any numerical error and
    monitoring the model during training and finding the relevance of every
    layer/parameter on the DNN outcome. We have collected a benchmark
    containing 40 buggy models and patches that contain real errors in
    deep learning applications from Stack Overflow and GitHub. Our benchmark
    can be used to evaluate automated debugging tools and repair techniques.
    We have evaluated our approach using this DNN bug-and-patch benchmark,
    and the results showed that our approach is much more effective than
    the existing debugging approach used in the state-of-the-practice
    Keras library. For 34/40 cases, our approach was able to detect faults
    whereas the best debugging approach provided by Keras detected 32/40
    faults. Our approach was able to localize 21/40 bugs whereas Keras
    did not localize any faults.
  }
}