PrivPkt

Privacy Preserving Collaborative Encrypted Network Traffic Classification

Interconnecting the following works:

• Differential Privacy
• Federated Learning (We plan to add split learning)
• Membership Inference Attacks
• Encrypted Traffic Classification

---

Federated Learning

We utilize Federated Averaging to enable the collaborative learning setting.

Ref: https://arxiv.org/abs/1602.05629

Differential Privacy

We make use of DPSGD to ensure a ceratin level of privacy.

Ref:https://arxiv.org/abs/1602.05629

Membership Inference Attacks

We make use of Shokri et al. Membership Inference Attacks to evaluate our mitigations.

Ref: https://arxiv.org/abs/1610.05820

Encrypted Traffic Classification

We tackle the problem of Encrypted Traffic Classification. We utilize the work of DeepPacket and use the ISCX Vpn 2016 Dataset to evaluate our work.

Ref: https://arxiv.org/abs/1709.02656

Ref: https://www.unb.ca/cic/datasets/vpn.html