Master Thesis 30HP – Dataset Drift & Dataset Poisoning Detection for Machine Learning Applications in Radar Warner Receivers
Saab Electronic Warfare and Aircraft Systems (EWAS) develop a wide range of systems, including passive receivers of radar signals such as Radar Warning Receivers (RWR). An RWR is responsible for detecting active radars and signaling when anything dangerous is approaching the aircraft.
An RWR can be operational for decades. During the lifecycle the signal environment will likely change, which the system must adapt to in order to stay relevant. An AI/ML-based function in an RWR will thus likely need to be updated on a regular basis, as new data is recorded.
An AI/ML-based function depends mainly on two factors: the software for training and running inference, and the dataset that the model is trained on. As the training and validation dataset grows over time, ideally the performance on the old validation dataset should be maintained or improved, while performance on new validation datasets should also be sufficient.
You will belong to the department Future Technology, which is a very competent and cross-functional team. We work with additive manufacturing (AM), machine learning (AI /ML), Business Intelligence (BI) etc. which creates a very innovative work environment.
Description of the master thesis
Your task in this master thesis will be to investigate methods related to any of the following topics, in the context of an AI/ML-model in an RWR:
- Dataset poisoning: A malicious actor could emit carefully designed signals, which if introduced into the training dataset, could increase the error rate in an AI/ML model.
- Dataset drift: Significant change in emitter behavior could increase the error rate in an AI/ML model, and could indicate that it is time for an update.
Specifically, given a dataset where some samples have labels, and a (potentially unlabeled) dataset which is a candidate for being introduced into the first dataset, define a metric or model that signals when the new dataset contain anomalies, significant drift, or potentially poisonous samples.
The AI/ML-model development will not be of focus in this master thesis, but rather the dataset lifecycle: How should new (potentially unlabeled) data be incorporated into the training and validation dataset?
We are looking for a student who:
- Is at the end of his/her master’s degree in Computer Science, Information Technology, Applied Physics, Electrical Engineering, or similar.
- Has an interest in data science and machine learning.
- Has taken courses related to data science, machine learning, deep learning.
Start January 2023, or according to agreement.
This position requires that you pass a security vetting based on the current regulations around/of security protection. For positions requiring security clearance additional obligations on citizenship may apply.
What you will be a part of
Saab is a leading defence and security company with an enduring mission, to help nations keep their people and society safe. Empowered by its 18,000 talented people, Saab constantly pushes the boundaries of technology to create a safer, more sustainable and more equitable world.
Business area Surveillance offers world-leading sensor technology in monitoring and decision support to protect against threats. The portfolio covers airborne, ground-based and naval radar, electronic warfare, C4I solutions, aviation systems and cyber security.
Last application date: 2022-12-15
Eric Norgren: firstname.lastname@example.org
Alexander Karlsson: email@example.com
If you aspire to help create and innovate whilst developing yourself in a learning culture, Saab may well have the perfect conditions for you to grow. We pride ourselves on a nurturing environment, where everyone is different yet we share the same goal – to keep people and society safe.