Demet Demir
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Demet Demir has a total of 2 co-authors. Here are the top co-authors by number of shared publications:
- Aysu Betin-Can
Graduate School of Informatics, Middle East Technical University, Ankara, Türkiye
2 shared publications - Elif Surer
Graduate School of Informatics, Middle East Technical University Ankara Türkiye
2 shared publications
Latest Publications
- Open Access
- 11 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 678-690
Test Selection for Deep Neural Networks using Meta-Models with Uncertainty Metrics
With the use of Deep Learning (DL) in safety-critical domains, the systematic testing of these systems has become a critical issue for human life. Due to the data-driven nature of Deep Neural Networks (DNNs), the effectiveness of tests is closely related to the adequacy of test datasets. Test data need to be labeled, which requires manual human effort and sometimes expert knowledge. DL system testers aim to select the test data that will be most helpful in identifying the weaknesses of the DNN model by using resources efficiently. To help achieve this goal, we propose a test data prioritization approach based on using a meta-model that gets uncertainty metrics as input, which are derived from outputs of other base models. Integrating different uncertainty metrics helps overcome individual limitations of these metrics and be effective in a wider range of scenarios. We train the meta-models with the objective of predicting whether a test input will lead the tested model to make an incorrect prediction or not. We conducted an experimental evaluation with popular image classification datasets and DNN models to evaluate the proposed approach. The results of the experiments demonstrate that our approach effectively prioritizes the test datasets and outperforms existing state-of-the-art test prioritization methods used in comparison. In the experiments, we evaluated the test prioritization approach from a distribution-aware perspective by generating test datasets with and without out-of-distribution data.
- 11 September 2024
- Published by Association for Computing Machinery (ACM)
- p. 678-690
Open AccessTest Selection for Deep Neural Networks using Meta-Models with Uncertainty Metrics
With the use of Deep Learning (DL) in safety-critical domains, the systematic testing of these systems has become a critical issue for human life. Due to the data-driven nature of Deep Neural Networks (DNNs), the effectiveness of tests is closely related to the adequacy of test datasets. Test data need to be labeled, which requires manual human effort and sometimes expert knowledge. DL system testers aim to select the test data that will be most helpful in identifying the weaknesses of the DNN model by using resources efficiently. To help achieve this goal, we propose a test data prioritization approach based on using a meta-model that gets uncertainty metrics as input, which are derived from outputs of other base models. Integrating different uncertainty metrics helps overcome individual limitations of these metrics and be effective in a wider range of scenarios. We train the meta-models with the objective of predicting whether a test input will lead the tested model to make an incorrect prediction or not. We conducted an experimental evaluation with popular image classification datasets and DNN models to evaluate the proposed approach. The results of the experiments demonstrate that our approach effectively prioritizes the test datasets and outperforms existing state-of-the-art test prioritization methods used in comparison. In the experiments, we evaluated the test prioritization approach from a distribution-aware perspective by generating test datasets with and without out-of-distribution data.
- Open Access
- 26 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 119481-119505
Distribution Aware Testing Framework for Deep Neural Networks
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.
- 26 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 119481-119505
Open AccessDistribution Aware Testing Framework for Deep Neural Networks
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.