PAC Learning of Arbiter PUFs
This chapter is based on Ganji et al., (J Cryptogr Eng Spec Sect Proofs, 2014:1–10, 2016, ), slightly modified to fit within the structure of this thesis. The current chapter aims at establishing a new representation of Arbiter PUFs that reflects the physical properties of these PUFs. This representation enables us to come up with new results on the learnability of Arbiter PUFs for given levels of accuracy and final model delivery confidence. This is in contrast to previous studies (e.g., Rührmair et al., Proceedings of the 17th ACM Conference on Computer and Communications Security. pp. 237–249, 2010, ), where it is not clear whether after the learning phase, a model of the Arbiter PUF with the desired level of accuracy would be delivered by the machine learner. Finally, we discuss the importance of our framework from the practical point of view.