Non-interactive Fully Encrypted Machine Learning Protocol for Inference¶
Year: 2024
type: scheme collection: ppml status: auto
Summary¶
Auto-extracted summary (unreviewed)
Proposes a functional-encryption-based PPML protocol in which a new FE scheme for quadratic polynomials has an encryption algorithm that is linear in the message (LinEnc), so encryptions can be composed with quadratic functions to evaluate arbitrary functions fully encrypted without any client interaction during inference. The protocol is proven secure in the malicious model and is demonstrated on a 2-layer neural network with quadratic activations, but the DCR-based implementation is slow: at 128-bit security, functional key generation and evaluation take 0.55 to 4 hours for input dimensions 1 to 6.
Abstract¶
As privacy concerns have arisen in machine learning, privacy-preserving machine learning (PPML) has received significant attention. Fully homomorphic encryption (FHE) and secure multi-party computation (MPC) are representative building blocks for PPML. However, in PPML protocols based on FHE and MPC, interaction between the client (who provides encrypted input data) and the evaluator (who performs the computation) is essential to obtain the final result in plaintext. Functional encryption (FE) is a promising candidate to remove this constraint, but existing FE-based PPML protocols are restricted to evaluating only simple ML models, such as one-layer neural networks, or they support partially encrypted PPML, which makes them vulnerable to information leakage beyond the inference results. In this paper, we propose a fully encrypted FE-based PPML protocol, which supports the evaluation of arbitrary functions over encrypted data with no information leakage during computation, for the first time. To achieve this, we introduce a new functional encryption scheme for quadratic polynomials, in which the encryption algorithm is realized as a linear function with respect to a message. This structural property allows for the composition of the encryption algorithm with arbitrary quadratic functions, thereby enabling multiple compositions of quadratic polynomials to compute arbitrary complex functions in an encrypted manner. Our FE-based PPML protocol is secure in the malicious model, which means that an adversary cannot obtain any information about the input data even though they intentionally deviate from the protocol. We then show how to use our protocol to build a fully encrypted 2-layer neural network model with quadratic activation functions and present experimental results.
Source: pdf
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | LinEnc functional encryption notion | llm | 0.85 |
| contribution | FE scheme for quadratic polynomials with message-linear encryption | llm | 0.85 |
| contribution | first fully encrypted FE-based PPML protocol supporting arbitrary functions | llm | 0.85 |
| contribution | malicious-model security with no intermediate leakage and no client involvement during inference | llm | 0.85 |
| contribution | 2-layer quadratic-activation neural network implementation and benchmarks | llm | 0.85 |
| schemes | functional encryption | llm | 0.85 |
| schemes | inner-product functional encryption (DCR-based) | llm | 0.85 |
| libraries | Charm | llm | 0.85 |
| models | 2-layer neural network with quadratic activations | llm | 0.85 |
| tasks | private inference | llm | 0.85 |
| tasks | classification | llm | 0.85 |
| hardware | Intel Xeon Silver 4208 CPU @ 2.10GHz, 30GB RAM | llm | 0.85 |
| security level | 128-bit (3072-bit primes p, q) | llm | 0.85 |
| threat model | malicious evaluator; client, evaluator and trusted key distributor; non-interactive inference | llm | 0.85 |
| interaction | other | llm | 0.85 |
| code url | https://github.com/swanhong/fully-encrypted-ml/.For (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/swanhong/fully-encrypted-ml (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/swanhong/fully-encrypted-ml | llm | 0.85 |
| limitations | significant operation time due to exponentiations in a large DCR group | llm | 0.85 |
| limitations | exponentiation is roughly 10^6 times more expensive than multiplication | llm | 0.85 |
| limitations | benchmarks only with toy parameters and small dimensions | llm | 0.85 |
| llm relevance | low | llm | 0.85 |
| key result | KeyGen time: 0.55h to 4.04h (two-layer model, dimension 1 to 6, 128-bit security) | llm | 0.85 |
| key result | evaluation (Dec) time: 0.58h to 4.02h (two-layer model, dimension 1 to 6, 128-bit security) | llm | 0.85 |
| key result | client encryption time: 0.70s to 1.11s (dimension 1 to 6) | llm | 0.85 |
| key result | setup time: 0.69s to 1.74s (dimension 1 to 6, excluding prime generation) | llm | 0.85 |
| key result | group exponentiation cost: 321 ms (Z_{N^2} with 3072-bit primes) | llm | 0.85 |
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: Functional encryption 68 [p.1], MPC 14 [p.1]
- library: ABY / ABY3 1 [p.4], EzPC / CrypTFlow 1 [p.4]
- model: Logistic regression 1 [p.23]
- dataset: Iris 5 [p.21], UCI repository 5 [p.3], Breast cancer (Wisconsin) 4 [p.3]
- hardware: Intel Xeon 1 [p.18]
- technique: Malicious model 10 [p.1], Semi-honest model 7 [p.4], Non-interactive 3 [p.1]
- assumption: Composite residuosity 26 [p.5], DDH 5 [p.26], 128-bit security 4 [p.18]
- task: Medical / healthcare 7 [p.1], Genomics 2 [p.1]
- topic: Privacy-preserving machine learning 67 [p.1], Homomorphic encryption 14 [p.1], Benchmark 8 [p.18], Private / secure inference 5 [p.5], Open source 2 [p.3]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/ppml/inference/hong-2024_non-interactive-fully-encrypted-ml-protocol-for-inference.pdf |
ppml/inference | 38 | 879.5 KB | 62b01a530119 |
was Hong et al. - 2024 - Non-interactive Fully Encrypted Machine Learning Protocol for Inference.pdf |
Database key: non-interactive-fully-encrypted-machine-learning-protocol-for-inference · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set non-interactive-fully-encrypted-machine-learning-protocol-for-inference COLUMN VALUE