WP 3 – Testing of Machine Learning Algorithms and Data

The significance of this project is its contribution towards reliable data-driven decision making. This, in turn, will contribute to the global aim of the AIDA project for enabling trustworthy data-driven real-time industrial IoT applications. Another significance and contribution (nationally and internationally) of the project would be in ML, testing, and IoT systems’ reliability. The primary aim is to design, develop, and deploy a new testing infrastructure that promotes data-driven IoT systems’ trustworthiness. The testing infrastructure will consider the continuous evolution of the system and the unpredictable behavior of the ML algorithms. Having such a testing infrastructure will prevent the system from storing noisy and corrupted data, avoid the production of inaccurate or imprecise decision results, and ensure the system’s quality while evolving over time.

To meet this aim and address all aforementioned research questions, we will introduce three activities in this project:

  • We conduct research on the data to introduce testing strategies for both the gathering and the training data.
  • Find the best real-time test generation strategy for ML correctness and an automated test oracle that verifies those generated test cases.
  • Finding out how to generate effective test cases that can tackle the internal interaction of neurons for deep learning algorithms.