Context-Aware Machine Learning in the Presence of Complex Noisy Data

  • Self Funded
  • York, England
  • Posted 9 months ago
  • Deadline: Open all year round

University of York

Department of Computer Science

Project Description

Multi-core timing analysis is one of the biggest challenges in Computer Science and especially in real-time systems. Multi-core timing analysis is not determining the Worst-Case Execution Time (WCET) of a software task [1] but instead building models of how much contention over shared resources can increase execution times [2]. Appropriate solutions are also needed soon by industry. The academic challenges come from the breadth of knowledge needed to address the problem.

Detailed knowledge of computer architectures, real-time systems and software testing need to be applied to statistical analysis and machine learning. More precisely blindly applying current machine learning techniques will not be successful not least as understanding the validity of the approaches is as appropriate for certification as is an accurate results [4]. Equally significant attention needs to be given that the data from testing addresses significant challenges from [3] including that the data and subsequent models are both sufficiently accurate and representative.

The proposed research would address the following using a test rig of Raspberry PI3s and real software from a Rolls-Royce aircraft engine.

1. Development of appropriate systematic software testing techniques that give confidence in the results building on the principles in [1] and [2] but deleting with the key challenges in [3].

2. Combined statistical and machine learning techniques that select the data to be monitoring and make accurate predictions of the execution time inflation based on contention over shared processing resources [5].

3. Explanations of the inaccuracies in the predictions and guidance for further testing [3,4,5].

4. Techniques for establishing the validity of both the noisy data and the resulting predictions [5, 6].

5. Schedulability analysis techniques based on the results of machine learning instead of single values for the WCET analysis [7].

Funding Information

This is a self funded project, there is no funding attached to it.

Application Process

Apply online.


[1] S. Law, I. Bate, Achieving Appropriate Test Coverage for Reliable Measurement-Based Timing Analysis, Euromicro Conference on Real-Time Systems, 2016.
[2] D. Griffin, B. Lesage, I. Bate, F. Soboczenski, R. Davis, Forecast-based interference: modelling multicore interference from observable factors, 25th International Conference on Real-Time Networks and Systems, (RTNS), 2017.
[3] S. Gil, I. Bate, G. Lima, L. Santinelli, A. Gogonel, L. Cucu-Grosjean, Open Challenges for Probabilistic Measurement-Based Worst-Case Execution Time, Embedded Systems Letters, 2017.
[4] S. Law, I. Bate, B. Lesage, Justifying the Service Provided to Low Criticality Tasks in a Mixed Criticality System, 28th International Conference on Real-Time Networks and Systems, (RTNS), 2020.
[5] X. Fang I. Bate, An Improved Sensor Calibration with Anomaly Detection and Removal, Sensors and Actuators B: Chemical, To Appear.
[6] I. Bate, D. Griffin, B. Lesage, Establishing Confidence and Understanding Uncertainty in Real-Time Systems, 28th International Conference on Real-Time Networks and Systems, (RTNS), 2020.
[7] M. Bartlett, I. Bate, J. Cussens, Learning Bayesian networks for improved instruction cache analysis, 9th International Conference on Machine Learning and Applications, 2010.

To apply for this PhD, please use the following application link: