Machine Learning approaches for traffic forecasting are increasingly being explored to improve resource allocation in mobile networks. However, these models remain vulnerable to adversarial manipulation.

In this presentation, we will examine the temporal constraints inherent to attacks on online time-series forecasting models: input values are shared across input sequences, the past is immutable, and future traffic values are unknown.

We will then present an attack on a wireless network capacity forecasting model that is able to significantly reduce the model’s accuracy while operating under these constraints. In particular, we discuss the broader impact of the attack and analyze which parameters specific to the case study influence its success.

About Louis Miermont

Louis Miermont is a PhD student at the IMDEA Networks Institute. He is part of the Cybersecurity Group under the supervision of Dr. Guillermo Suarez-Tangil. He received his Master’s degree from the ESIEA Graduate School of Engineering, and was previously working as a Cybersecurity Analyst for Airbus, investigating security incidents targeting external companies of “vital importance”. His research interests currently lie at the intersection between Artificial Intelligence and Cybersecurity.

This event will be conducted in English