FBAN: A Fully Homomorphic Encryption Compatible Bottleneck Attention Network for Privacy-Preserving Behavioral Authentication
FBAN enables TFHE-based continuous authentication from touch behavior at 128-bit security.
The paper proposes FBAN, a TFHE-compatible Bottleneck Attention Network for continuous authentication that keeps behavioral features encrypted during inference. A float-then-quantization-aware pipeline compiles the model into TFHE circuits, and a client-server protocol uses session-bound blinding so the server never sees raw touch dynamics. The security analysis targets an honest-but-curious server under TFHE IND-CPA and claims resistance to replay and impersonation without the secret key. On two public touchscreen datasets, encrypted inference retains strong utility with a lightweight model at at least 128-bit security.
- Protects touch behavioral features during inference, not only at rest.
- Integer pipeline uses pretraining, quantization-aware training, and TFHE compilation.
- Session-bound blinding keeps raw features hidden from the server.
- Tested on two public touchscreen datasets at least 128-bit security.
From the original article
1 min read at arxiv.orgContinuous authentication (CA) strengthens session security by repeatedly verifying the user during device interaction, yet it inherently relies on highly sensitive behavioral traces (e. g. , fine-grained touch dynamics) that are often outsourced to cloud/edge services for scalable inference.
By Jichao Xiong · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.