skip to main content

Explore Research Products in the NSF-PAR

Title: UMLAUT: Debugging Deep Learning Programs using Program Structure and Model Behavior
Training deep neural networks can generate non-descriptive error messages or produce unusual output without any explicit errors at all. While experts rely on tacit knowledge to apply debugging strategies, non-experts lack the experience required to interpret model output and correct Deep Learning (DL) programs. In this work, we identify DL debugging heuristics and strategies used by experts, andIn this work, we categorize the types of errors novices run into when writing ML code, and map them onto opportunities where tools could help. We use them to guide the design of Umlaut. Umlaut checks DL program structure and model behavior against these heuristics; provides human-readable error messages to users; and annotates erroneous model output to facilitate error correction. Umlaut links code, model output, and tutorial-driven error messages in a single interface. We evaluated Umlaut in a study with 15 participants to determine its effectiveness in helping developers find and fix errors in their DL programs. Participants using Umlaut found and fixed significantly more bugs and were able to implement fixes for more bugs compared to a baseline condition.  more » « less
Award ID(s):
1955394
NSF-PAR ID:
10309311
Author(s) / Creator(s):
 ;  ;  
Date Published:
Journal Name:
CHI Conference on Human Factors in Computing Systems (CHI '21)
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ( , 14th Innovations in Theoretical Computer Science Conference, {ITCS} 2023)
    Kaiai, Yael Tauman (Ed.)
    Societal accumulation of knowledge is a complex process. The correctness of new units of knowledge depends not only on the correctness of new reasoning, but also on the correctness of old units that the new one builds on. The errors in such accumulation processes are often remedied by error correction and detection heuristics. Motivating examples include the scientific process based on scientific publications, and software development based on libraries of code. Natural processes that aim to keep errors under control, such as peer review in scientific publications, and testing and debugging in software development, would typically check existing pieces of knowledge - both for the reasoning that generated them and the previous facts they rely on. In this work, we present a simple process that models such accumulation of knowledge and study the persistence (or lack thereof) of errors. We consider a simple probabilistic model for the generation of new units of knowledge based on the preferential attachment growth model, which additionally allows for errors. Furthermore, the process includes checks aimed at catching these errors. We investigate when effects of errors persist forever in the system (with positive probability) and when they get rooted out completely by the checking process. The two basic parameters associated with the checking process are the probability of conducting a check and the depth of the check. We show that errors are rooted out if checks are sufficiently frequent and sufficiently deep. In contrast, shallow or infrequent checks are insufficient to root out errors. 
    more » « less
  2. ; ; ( , 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE))
    null (Ed.)
    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. 
    more » « less
  3. ; ; ; ( , Zenodo)

    Efficiency is essential to support responsiveness w.r.t. ever-growing datasets, especially for Deep Learning (DL) systems. DL frameworks have traditionally embraced deferred execution-style DL code that supports symbolic, graph-based Deep Neural Network (DNN) computation. While scalable, such development tends to produce DL code that is error-prone, non-intuitive, and difficult to debug. Consequently, more natural, less error-prone imperative DL frameworks encouraging eager execution have emerged but at the expense of run-time performance. While hybrid approaches aim for the "best of both worlds," the challenges in applying them in the real world are largely unknown. We conduct a data-driven analysis of challenges—and resultant bugs—involved in writing reliable yet performant imperative DL code by studying 250 open-source projects, consisting of 19.7 MLOC, along with 470 and 446 manually examined code patches and bug reports, respectively. The results indicate that hybridization: (i) is prone to API misuse, (ii) can result in performance degradation—the opposite of its intention, and (iii) has limited application due to execution mode incompatibility. We put forth several recommendations, best practices, and anti-patterns for effectively hybridizing imperative DL code, potentially benefiting DL practitioners, API designers, tool developers, and educators.

    Support for this project was provided by PSC-CUNY Award #638010051, jointly funded by The Professional Staff Congress and The City University of New York. 
    more » « less
  4. ; ; ; ( , International Conference on Mining Software Repositories)
    Efficiency is essential to support responsiveness w.r.t. ever-growing datasets, especially for Deep Learning (DL) systems. DL frameworks have traditionally embraced deferred execution-style DL code that supports symbolic, graph-based Deep Neural Network (DNN) computation. While scalable, such development tends to produce DL code that is error-prone, non-intuitive, and difficult to debug. Consequently, more natural, less error-prone imperative DL frameworks encouraging eager execution have emerged but at the expense of run-time performance. While hybrid approaches aim for the "best of both worlds," the challenges in applying them in the real world are largely unknown. We conduct a data-driven analysis of challenges---and resultant bugs---involved in writing reliable yet performant imperative DL code by studying 250 open-source projects, consisting of 19.7 MLOC, along with 470 and 446 manually examined code patches and bug reports, respectively. The results indicate that hybridization: (i) is prone to API misuse, (ii) can result in performance degradation---the opposite of its intention, and (iii) has limited application due to execution mode incompatibility. We put forth several recommendations, best practices, and anti-patterns for effectively hybridizing imperative DL code, potentially benefiting DL practitioners, API designers, tool developers, and educators. 
    more » « less
  5. ; ; ( , The 27th ACM Conference on Innovation and Technology in Computer Science Education)
    There is an increasing need for knowledgeable K-12 computer science (CS) teachers. It is necessary to inform teachers how to debug and help their students debug programs. Research has shown that debugging is difficult for novices because the process requires different skills from creating programs and instructing students how to debug can help them acquire these skills. To this end, we developed a CS professional development for middle grade teachers (grades 5th-8th/ages 10-13) that includes lessons on debugging. The teachers completed debugging activities that involved finding bugs in Scratch programs and explaining how they would help their students in debugging. We qualitatively analyzed their responses and found that teachers successfully identified the problem but they struggled to locate it in the code. In considering how they would help students who had such a bug, the teachers often focused on helping the student find a solution for the bug rather than on identifying the problem or its source. Finally, teachers’ ability to identify bugs and the pedagogical strategies to engage students in this process differed based on CS teaching experience and prior CS knowledge. This work contributes to our understanding of teachers’ debugging abilities and advances our knowledge on how to support teachers in teaching their students how to debug their programs. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email