Comparing Input Prioritization Techniques for Testing Deep Learning Algorithms

Mosin, Vasilii; Staron, Miroslaw; Durisic, Darko; Neto, Francisco Gomes de Oliveira; Pandey, Sushant Kumar; Koppisetty, Ashok Chaitanya

doi:10.1109/seaa56994.2022.00020

Cited by 3 publications

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…APFD metric is used to compare the fault detection speed of test selection methods and metrics [26], [28] [83]. The range of APFD is between 0 and 1.…”

Section: B: Test Data Prioritizationmentioning

confidence: 99%

Distribution Aware Testing Framework for Deep Neural Networks

Demir,

Can,

Surer

2023

IEEE Access

View full text Add to dashboard Cite

The increasing use of deep learning (DL) in safety-critical applications highlights the critical need for systematic and effective testing to ensure system reliability and quality. In this context, researchers have conducted various DL testing studies to identify weaknesses in Deep Neural Network (DNN) models, including exploring test coverage, generating challenging test inputs, and test selection. In this study, we propose a generic DNN testing framework that takes into consideration the distribution of test data and prioritizes them based on their potential to cause incorrect predictions by the tested DNN model. We evaluated the proposed framework using the image classification as a use case. We conducted empirical evaluations by implementing each phase with carefully chosen methods. We employed Variational Autoencoders to identify and eliminate out-of-distribution data from the test datasets. Additionally, we prioritize test data that increase uncertainty in the model, as these cases are more likely to reveal potential faults. The elimination of out-of-distribution data enables a more focused analysis to uncover the sources of DNN failures while using prioritized test data reduces the cost of test data labeling. Furthermore, we explored the use of post-hoc explainability methods to identify the cause of incorrect predictions, a process similar to debugging. This study can be a prelude to incorporating explainability methods into the model development process after testing.

show abstract

“…APFD metric is used to compare the fault detection speed of test selection methods and metrics [26], [28] [83]. The range of APFD is between 0 and 1.…”

Section: B: Test Data Prioritizationmentioning

confidence: 99%

Distribution Aware Testing Framework for Deep Neural Networks

Demir,

Can,

Surer

2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Even seemingly routine tasks such as splitting data into training, testing, and validation sets must be carefully pre-planned. The data collection process can be time-consuming and resource-intensive, ensuring the accuracy of ML models [8] depends on the quality and correctness of the data.…”

Section: Introductionmentioning

confidence: 99%

Data Handling for Assuring Production Quality of Image Intensive Autonomous Drive Systems: An Industrial Case Study

Pandey,

Mosin,

Durisic

et al. 2023

JSEAS

View full text Add to dashboard Cite

Automotive software uses new Machine Learning (ML) algorithms in increased number of systems, including active safety ones. However, with this new paradigm, new challenges arise in the domain of safety-critical automotive software. This article reports on a case study of the development of ML-based vision perception systems at one vehicle Original Equipment Manufacturer (OEM). We investigate how image-intensive perception systems are developed, both from the perspective of ML development processes and automotive software development processes. We conducted interviews with four engineers who were involved in the development process and subsequently performed thematic coding to extract key findings. We focus on how the teams are involved in the process of assuring the production quality of the Society of Automotive Engineers (SAE) level 3 functionality in modern passenger cars. We examine how the ML development process phases (e.g., data collection, model training, and model validation) align with the automotive software development phases (prototype development, software development, validation, deployment). The study found that the development process for ML-based vision perception systems in active safety allows for flexibility to adapt to changes in data collection, and integrates ML model development into the software development process. The investigated approach combines the Agile SAFe model, ML-model development, and the standard automotive V-process model. This study shows an example of combining these development processes using an industrial case study and presents the essential alignment points between different phases of these processes. The study also recommends best practices for developing similar systems in other companies.

show abstract