Research map / Federated learning and privacy / Privacy attacks and secure learning
Low Latency Privacy Preserving Inference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use varied encrypted-data representations and change representation across layers instead of using one SIMD representation | LoLa on MNIST · MNIST | latency: 24.8 -> 2.2 | Helped |
| Use varied encrypted-data representations and change representation across layers instead of using one SIMD representation | LoLa on MNIST · MNIST | accuracy: 98.95% -> 98.95% | No clear effect |
| Use a dense input representation and convert it to convolutional representation (ablation) | LoLa-Dense · MNIST | latency: 2.2 -> 7.2 | Hurt |
| Use a smaller network with one convolution layer, square activation, and dense layer (ablation) | LoLa-Small · MNIST | latency: 2.2 -> 0.29 | Helped |
| Use a smaller network with one convolution layer, square activation, and dense layer (ablation) | LoLa-Small · MNIST | accuracy: 98.95% -> 96.92% | Hurt |
| Encode input with convolution representation and use LoLa operations rather than per-node SIMD messages | LoLa on CIFAR-10 · CIFAR-10 | memory feasibility: 100's of Gigabytes -> 12 GB | Helped |
| Use convolutional/interleaved representation changes for the encrypted CIFAR-10 network | LoLa on CIFAR-10 · CIFAR-10 | latency: more than 6 hours -> 730 seconds | Helped |
| Compute AlexNet deep features before encryption and use a linear predictor on encrypted features | Private inference with transfer learning · CalTech-101 | encrypted inference time: 0.16 seconds | Helped |
| Use a linear classifier rather than more complex classifiers on AlexNet deep features (ablation) | Private inference with transfer learning · CalTech-101 | accuracy | No clear effect |
Closest papers by meaning
- SAFENet: A Secure, Accurate and Fast Neural Network Inference (ICLR 2021 · similarity 0.88)
- SpENCNN: Orchestrating Encoding and Sparsity for Fast Homomorphically Encrypted Neural Network Inference (ICML 2023 · similarity 0.88)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.85)
- Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption (ICML 2024 · similarity 0.84)
- Backdoor Channels Hidden in Latent Space: Cryptographic Undetectability in Modern Neural Networks (NeurIPS 2026 · similarity 0.84)
- PrivCirNet: Efficient Private Inference via Block Circulant Transformation (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.