Research map / Federated learning and privacy / Privacy attacks and secure learning
PCFormer: Accelerating Privacy-preserving Transformer Inference by Partition and Combination
Bo Zeng, Zhi Pang, Yuyang Zhang, Kai Zhao, Tian Wu, Geying Yang, Lina Wang, Run Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 partition-and-combination of Softmax and GeLU instead of CipherPrune pruning | private BERT inference · BERT-Large (LAN) | end-to-end time (s): 210.9 -> 110.6 | Improved |
| use partition-and-combination instead of CipherPrune | private BERT inference · BERT-Large (LAN) | communication (GB): 25.3 -> 13.2 | Improved |
| use PCFormer instead of Bolt | private BERT inference · BERT-Base (LAN) | end-to-end time (s): 260.5 -> 62.2 | Improved |
| use PCFormer instead of Bolt or CipherPrune | BERT-Base · MNLI/QNLI/SST-2/MRPC | accuracy: MNLI 84.73 vs 84.71 (Bolt) | No clear effect |
| apply partition and combination to ReLU structure | linear-nonlinear block · ReLU | total latency (ms): 774.3 -> 257.1 | Improved |
| apply partition and combination to SiLU structure | linear-nonlinear block · SiLU | total latency (ms): 3950.5 -> 1033.3 | Improved |
| raise token sparsity weight lambda_tok above 0.05 (ablation) | PCFormer fine-tuning · BERT-Base | accuracy | Hurt |
| keep lambda_tok and lambda_chn below 0.05 (ablation) | PCFormer fine-tuning · BERT-Base | accuracy | No clear effect |
| vary token sparsity versus channel sparsity (ablation) | PCFormer fine-tuning · BERT-Base | accuracy | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CipherPrune — CipherPrune: Efficient and Scalable Private Transformer Inference (ICLR 2025) · better on 10 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption (ICML 2024 · similarity 0.89)
- CipherPrune: Efficient and Scalable Private Transformer Inference (ICLR 2025 · similarity 0.89)
- Encryption-Friendly LLM Architecture (ICLR 2025 · similarity 0.88)
- Nimbus: Secure and Efficient Two-Party Inference for Transformers (NeurIPS 2024 · similarity 0.87)
- United We Stand: Accelerating Privacy-Preserving Neural Inference by Conjunctive Optimization with Interleaved Nexus (AAAI 2024 · similarity 0.87)
- Ditto: Quantization-aware Secure Inference of Transformers upon MPC (ICML 2024 · similarity 0.86)
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.
