Deep learning models play a crucial role in important sectors like healthcare, finance, and recruitment. However, these models can harbor biases and unfairness that may negatively impact those who rely on them. While there are algorithms designed to improve fairness, the resilience of these models to adversarial attacks, particularly the emerging threat of Trojan (or backdoor) attacks, has not been thoroughly examined. To address this gap, we introduce TrojFair, a Trojan fairness attack methodology. TrojFair creates a model that performs accurately and fairly under normal conditions but, when triggered, discriminates and produces incorrect results for specific groups. This type of attack is especially stealthy and dangerous as it can bypass existing fairness and trojan detection methods, appearing fair during regular use. Our results show that TrojFair achieves a success rate of over 88.77% in targeting specific groups, with a minimal average accuracy loss of less than 0.44%. Additionally, it consistently demonstrates a significant bias score, distinguishing between targeted and non-targeted groups across various datasets and model architectures.
Fully homomorphic encryption (FHE) protects data privacy in cloud computing by enabling computations to directly occur on cipher-texts. To improve the time-consuming FHE operations, we present an electro-optical (EO) FHE accelerator, CryptoLight. Compared to prior FHE accelerators, on average, CryptoLight reduces the latency of various FHE applications by >94.4% and the energy by >95%.
This paper presents OFHE, an electro-optical accelerator designed to process Discretized TFHE (DTFHE) operations, which encrypt multi-bit messages and support homomorphic multiplications, lookup table operations and full-domain functional bootstrappings. While DTFHE is more efficient and versatile than other fully homomorphic encryption schemes, it requires 32-, 64-, and 128-bit polynomial multiplications, which can be time-consuming. Existing TFHE accelerators are not easily upgradable to support DTFHE operations due to limited datapaths, a lack of datapath bit-width reconfigurability, and power inefficiencies when processing FFT and inverse FFT (IFFT) kernels. Compared to prior TFHE accelerators, OFHE addresses these challenges by improving the DTFHE operation latency by 8.7%, the DTFHE operation throughput by 57%, and the DTFHE operation throughput per Watt by 94%.