Abstract:Homomorphic encryption (HE) is a promising technique used for privacy-preserving computation. Since HE schemes only support primitive polynomial operations, homomorphic evaluation of polynomial approximations for non-polynomial functions plays an important role in privacy-preserving machine learning. In this paper, we introduce a simple solution to approximating any functions, which might be overmissed by researchers: just using the neural networks for regressions. By searching decent superparameters, neural networks can achieve near-optimal computation depth for a given function with fixed precision, thereby reducing the modulus consumed.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the challenges encountered in homomorphic evaluation of non - polynomial functions in Homomorphic Encryption (HE). Specifically, HE schemes only support addition and multiplication operations. Therefore, for circuit calculations involving non - polynomial functions, these functions must first be replaced with polynomial approximations. However, performing such approximations over large intervals will lead to an overly high order of the polynomial, thereby increasing the computational overhead. #### Main problems: 1. **Homomorphic evaluation of non - polynomial functions**: How to efficiently evaluate non - polynomial real - valued functions in the homomorphic - encrypted state. 2. **Polynomial approximation over large intervals**: When performing polynomial approximation over large intervals, how to reduce the computational complexity and modulus consumption. 3. **Minimizing multiplication depth**: How to minimize the multiplication depth in homomorphic calculations while ensuring a reasonable maximum error. #### Solutions in the paper: The authors propose a simple and effective solution, that is, using neural networks for regression to approximate arbitrary functions. By carefully adjusting the hyperparameters, the neural network can achieve a near - optimal computational depth at a given precision, thereby reducing modulus consumption. Specific methods include: - **Using neural networks for polynomial approximation**: Neural networks can be used to approximate arbitrary functions and can efficiently evaluate polynomials by choosing an appropriate architecture, reducing the multiplication depth. - **Applicable to large intervals**: This method can perform polynomial approximation over larger intervals, such as the evaluation of the Sigmoid function on the intervals [- 30, + 30], [- 50, + 50] and [- 70, + 70]. - **Optimizing computational depth**: By searching for suitable parameters, the neural network can evaluate the approximation of a given function on the interval [- R, + R] with a multiplication depth of O(logR), thereby reducing the computational overhead. #### Experimental results: Experiments show that this method can effectively approximate various functions and can achieve high precision at a reasonable modulus cost when evaluating the Sigmoid function over large intervals. In summary, the main objective of this paper is to provide a simple and efficient method for evaluating non - polynomial functions under homomorphic encryption using neural networks, especially suitable for cases where calculations need to be performed over large intervals.

A Simple Solution for Homomorphic Evaluation on Large Intervals

Accurate Low-Degree Polynomial Approximation of Non-polynomial Operators for Fast Private Inference in Homomorphic Encryption

Optimized Privacy-Preserving CNN Inference With Fully Homomorphic Encryption

Three-Input Ciphertext Multiplication for Homomorphic Encryption

Hyperdimensional Computing as a Rescue for Efficient Privacy-Preserving Machine Learning-as-a-Service

Fast and Accurate Homomorphic Softmax Evaluation

Approximate Methods for the Computation of Step Functions in Homomorphic Encryption.

Intelligent Beta-Based Polynomial Approximation of Activation Functions for a Robust Data Encryption System

MOFHEI: Model Optimizing Framework for Fast and Efficient Homomorphically Encrypted Neural Network Inference

Efficient Privacy-Preserving KAN Inference Using Homomorphic Encryption

HEQuant: Marrying Homomorphic Encryption and Quantization for Communication-Efficient Private Inference

A Pervasive, Efficient and Private Future: Realizing Privacy-Preserving Machine Learning Through Hybrid Homomorphic Encryption

Verifiable computation over encrypted data via MPC-in-the-head zero-knowledge proofs

When approximate design for fast homomorphic computation provides differential privacy guarantees

Optimized Layerwise Approximation for Efficient Private Inference on Fully Homomorphic Encryption

Self-learning activation functions to increase accuracy of privacy-preserving Convolutional Neural Networks with homomorphic encryption

A Multi-Layer Parallel Hardware Architecture for Homomorphic Computation in Machine Learning

Privacy Preserving Inference for Deep Neural Networks: Optimizing Homomorphic Encryption for Efficient and Secure Classification

A privacy preserving homomorphic computing toolkit for predictive computation

Efficient Privacy-Preserving Convolutional Spiking Neural Networks with FHE

Securer and Faster Privacy-Preserving Distributed Machine Learning