tinyML Talks local Webcast – recorded January 13, 2021
“Security of Edge AI against Hardware Attacks”
Matthias Probst and Manuel Brosch (TUM)
Neural networks are widespread in usage and are amenable too many applications. However, depending on the task a network should perform, training can be compute and time consuming. Consequently, in many cases a neural network is an intellectual property worthwhile to protect.
A possible attack target is to reverse engineer the network and build a copy of it. In the domain of edge AI, hardware attacks like side-channel analysis must be considered, since an attacker may get physical access to the device. In this talk the security of neural networks against hardware attacks will be discussed. Moreover, an insight into possible countermeasures is given.