A Graph-Based Analysis of Comparability, Scale, and Attribution in Private Transformer Inference

arXiv:2610.19607 · cs.CR, cs.LG · Submitted 2026-10-14 · Read on arXiv

cs.CR, cs.LG

Submitted: 2026-10-14

Updated: 2026-10-14

Comments: 21 pages, 10 figures, 11 tables

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Transformer — 1× in this paper · explained in 2 episodes

Sources

• A Survey on Private Transformer Inference
• Private Transformer Inference in MLaaS: A Survey
• PUMA: Secure Inference of LLaMA-7B in Five Minutes
• Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache
• SoK: Privacy-preserving Deep Learning with Homomorphic Encryption
• SoK: Can Fully Homomorphic Encryption Support General AI Computation? A Functional and Cost Analysis
• Towards Efficient Privacy-Preserving Machine Learning: A Systematic Review from Protocol, Model, and System Perspectives
• CryptoGen: Secure Transformer Generation with Encrypted KV-Cache Reuse
• AEGIS: Scaling Long-Sequence Homomorphic Encrypted Transformer Inference via Hybrid Parallelism on Multi-GPU Systems
• EncFormer: Secure and Efficient Transformer Inference over Encrypted Data
• Scaling up FHE-based Privacy-Preserving ML: Higher Throughput, Longer Inputs for LLama-3-8B
• TruncFormer: Private LLM Inference Using Only Truncations
• FuseFSS: Efficient Secure LLM Inference with Function Secret Sharing
• MPC-Minimized Secure LLM Inference
• Power-Softmax: Towards Secure LLM Inference over Encrypted Data
• ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour
• ENSI: Efficient Non-Interactive Secure Inference for Large Language Models
• Practical Secure Inference Algorithm for Fine-tuned Large Language Model Based on Fully Homomorphic Encryption
• LLMs Can Understand Encrypted Prompt: Towards Privacy-Computing Friendly Transformers
• Entropy-Guided Attention for Private LLMs

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Estimating the Social Cost of Corporate Data Breaches
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills