# CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption

Source: [CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption](<https://www.sota2.com/research/paper/698181e2a8b01e76cc8e28cf>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-12T05:00:04.826Z

Authors: Hanjun Park, Byeongseo Min, Jiheon Woo, Min-Wook Jeong, Jongho Shin, Yongwoo Lee, Young-Sik Kim, Yongjune Kim

Published: 2026-02-02T04:33:12.000Z

## Abstract

Homomorphic encryption (HE) is a prominent framework for privacy-preserving machine learning, enabling inference directly on encrypted data. However, evaluating softmax, a core component of transformer architectures, remains particularly challenging in HE due to its multivariate structure, the large dynamic range induced by exponential functions, and the costly division operation. In this paper, we propose CGF-softmax, which reformulates the softmax denominator through the cumulant generating function (CGF). By eliminating both homomorphic division and explicit maximum subtraction, this reformulation substantially reduces multiplicative depth while preserving key properties of softmax. Extensive experiments on Vision Transformers and large language models show that CGF-softmax provides an efficient and accurate approximation of softmax in encrypted inference. In particular, it achieves inference accuracy close to that of high-depth exact methods, while requiring substantially lower computational cost through reduced multiplicative depth.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [NLP Classification on (SST-2, WiC, PIQA, MNLI, ANLI, Social IQA, HellaSwag, ARC, RTE, MRPC, QNLI, Banking77, Clinc150)](<https://www.sota2.com/research/sota/nlp-classification-on-sst-2-wic-piqa-mnli-anli-social-iqa-hellaswag-arc-rte-mrpc-qnli-banking77-clinc150.md>): NLP Classification; NLP Benchmarks SST-2, WiC, PIQA, MNLI, ANLI, Social IQA, HellaSwag, ARC, RTE, MRPC, QNLI, Banking77, Clinc150; Model Depth: 89.

- [Image Classification on ImageNet-1k (Mult. Depth Reporting)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-mult-depth-reporting.md>): Image Classification; ImageNet-1K; Mult. Depth: 710.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2602.01621>)

[PDF](<https://arxiv.org/pdf/2602.01621.pdf>)