With the increasing application of deep learning (DL) models in many safety-critical scenarios, effective and efficient DL testing techniques are much in demand to improve the quality of DL models. One of the major challenges is the data gap between the training data to construct the models and the testing data to evaluate them. To bridge the gap, testers aims to collect an effective subset of inputs from the testing contexts, with limited labeling effort, for retraining DL models.

To assist the subset selection, we propose \textbf{M}ultiple-Boundary \textbf{C}lustering and \textbf{P}rioritization (\textbf{MCP}), a technique to cluster test samples into the boundary areas of multiple boundaries for DL models and specify the priority to select samples evenly from all boundary areas, to make sure enough useful samples for each boundary reconstruction.

To evaluate MCP, we conduct an extensive empirical study with three popular DL models and 33 simulated testing contexts. The experiment results show that, compared with state-of-the-art baseline methods, on effectiveness, our approach MCP has a significantly better performance by evaluating the improved quality of retrained DL models; on efficiency, MCP also has the advantages in time costs.

Weijun Shen

Nanjing University

Yanhui Li

Department of Computer Science and Technology, Nanjing University

Lin Chen

Nanjing University

China

YuanLei Han

Nanjing University

Yuming Zhou

Nanjing University

Baowen Xu

State Key Laboratory for Novel Software Technology, Nanjing University

Deep learning (DL) has recently started to be applied in many applications, e.g., autonomous driving, speech recognition, and natural language processing. Yet, many state-of-the-art DL systems are still vulnerable to adversarial examples, which hinders their adoptions in safety- and security-critical scenarios. While some recent progress has been made in analyzing the robustness of feed-forward neural networks, the robustness analysis for stateful DL systems, such as recurrent neural networks (RNNs), still remains largely uncharted. In this paper, we propose MARBLE, a model-based approach for quantitative robustness analysis of real-world RNN-based DL systems. MARBLE first profiles RNNs using training data to collect information on how models behave under controlled perturbations. We then build a probabilistic model to compactly characterize the behavioral robustness of RNNs, through abstraction. Furthermore, we propose a refinement algorithm to iteratively derive a precise abstraction which enables accurate quantification of the robustness measures. We evaluate the effectiveness of MARBLE on both LSTM and GRU models trained separately with three popular natural language datasets. The results demonstrate that (1) our refinement algorithm is more efficient in deriving an accurate abstraction than the random strategy, and (2) MARBLE enables quantitative robustness analysis, in rendering better efficiency, accuracy, and scalability than state-of-the-art techniques.

Xiaoning Du

Nanyang Technological University

Yi Li

Nanyang Technological University, Singapore

Singapore

Xiaofei Xie

Nanyang Technological University

Lei Ma

Kyushu University

Japan

Yang Liu

Nanyang Technological University, Singapore

Singapore

Jianjun Zhao

Kyushu University

Japan

Machine learning software is being used in many applications (finance, hiring, admissions, criminal justice) having a huge social impact. But sometimes the behavior of this software is biased and it shows discrimination based on some sensitive attributes such as sex, race etc. Prior works concentrated on finding and mitigating bias in ML models. A recent trend is using instance-based model agnostic explanation methods such as LIME[1] to find out bias in the model prediction. Our work concentrates on finding shortcomings of current bias measures and explanation methods. We show how our proposed method based on K nearest neighbors can overcome those shortcomings and find the underlying bias of black-box models. Our results are more trustworthy and helpful for the practitioners. Finally, We describe our future framework combining explanation and planning to build fair software

Link to Publication: https://arxiv.org/abs/2007.02893

Joymallya Chakraborty

North Carolina State University, USA

United States

Kewen Peng

North Carolina State University

Tim Menzies

North Carolina State University, USA

United States