Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| 格式: | Preprint |
| 出版: |
2026
|
| 主題: | |
| 在線閱讀: | https://arxiv.org/abs/2602.01621 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| _version_ | 1866914541626982400 |
|---|---|
| author | Park, Hanjun Min, Byeongseo Woo, Jiheon Jeong, Min-Wook Shin, Jongho Lee, Yongwoo Kim, Young-Sik Kim, Yongjune |
| author_facet | Park, Hanjun Min, Byeongseo Woo, Jiheon Jeong, Min-Wook Shin, Jongho Lee, Yongwoo Kim, Young-Sik Kim, Yongjune |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_01621 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption Park, Hanjun Min, Byeongseo Woo, Jiheon Jeong, Min-Wook Shin, Jongho Lee, Yongwoo Kim, Young-Sik Kim, Yongjune Cryptography and Security Machine Learning 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. |
| title | CGF-Softmax: A Cumulant-Based Softmax Reformulation for Efficient Inference under Homomorphic Encryption |
| topic | Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2602.01621 |