[MG1] Optical Networks for Computing

11:45 AM - 12:00 PM JST(2:45 AM - 3:00 AM UTC)

[MG1-3] Assessing the Robustness of In-Switch Neural Networks Against Adversarial DDoS Attacks

○Giulio Zingrillo^1,2,3, Emilio Paolini¹, Nicola Andriolli³, Francesco Paolucci², Filippo Cugini², Giampiero Contestabile¹, Piero Castoldi¹, Lorenzo De Marinis¹ (1. Scuola Superiore Sant'Anna (Italy), 2. CNIT - National Inter-University Consortium for Telecommunications (Italy), 3. University of Pisa (Italy))

Keywords:

Converged mobile and optical access networks,Fog/edge computing infrastructure and applications,Network function virtualization (NFV)

In data-plane deployment of DNNs allows for improved edge functionality and intrusion detection. We report on the increased resilience of in-switch LUT-distilled neural networks to white-box and black-box adversarial attacks.