Cryptographic inference for large language model via secret sharing

Inference services based on large language models may lead to the leakage of user input hints to the server or proprietary model weights to the user. Cryptographic techniques such as secure multi-party computation and homomorphic encryption provide feasible solutions to the above problems, but they...

Full description

Saved in:
Bibliographic Details
Main Authors: CHENG Ke, XIA Yuheng, DAI Chuanyun, FU Jiaxuan, ZHU Xinghui, SHEN Yulong
Format: Article
Language:zho
Published: Editorial Department of Journal on Communications 2025-06-01
Series:Tongxin xuebao
Subjects:
Online Access:http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025115/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849424504328028160
author CHENG Ke
XIA Yuheng
DAI Chuanyun
FU Jiaxuan
ZHU Xinghui
SHEN Yulong
author_facet CHENG Ke
XIA Yuheng
DAI Chuanyun
FU Jiaxuan
ZHU Xinghui
SHEN Yulong
author_sort CHENG Ke
collection DOAJ
description Inference services based on large language models may lead to the leakage of user input hints to the server or proprietary model weights to the user. Cryptographic techniques such as secure multi-party computation and homomorphic encryption provide feasible solutions to the above problems, but they are still difficult to apply practically to the task of inference over large language models due to the excessive computational and communication overhead. Based on this, a lightweight secret-sharing-based cryptographic inference scheme for large language models was proposed, by which inference could be performed efficiently and accurately while ensuring that neither user inputs nor model parameters were revealed. The experimental results show that the proposed scheme improves the efficiency by 1.2~10 times and reduces the communication cost by 20%~90% compared with the existing state-of-the-art works.
format Article
id doaj-art-e325f767d11e4e3d9a84644ad2bdec92
institution Kabale University
issn 1000-436X
language zho
publishDate 2025-06-01
publisher Editorial Department of Journal on Communications
record_format Article
series Tongxin xuebao
spelling doaj-art-e325f767d11e4e3d9a84644ad2bdec922025-08-20T03:30:09ZzhoEditorial Department of Journal on CommunicationsTongxin xuebao1000-436X2025-06-0146168184114258215Cryptographic inference for large language model via secret sharingCHENG KeXIA YuhengDAI ChuanyunFU JiaxuanZHU XinghuiSHEN YulongInference services based on large language models may lead to the leakage of user input hints to the server or proprietary model weights to the user. Cryptographic techniques such as secure multi-party computation and homomorphic encryption provide feasible solutions to the above problems, but they are still difficult to apply practically to the task of inference over large language models due to the excessive computational and communication overhead. Based on this, a lightweight secret-sharing-based cryptographic inference scheme for large language models was proposed, by which inference could be performed efficiently and accurately while ensuring that neither user inputs nor model parameters were revealed. The experimental results show that the proposed scheme improves the efficiency by 1.2~10 times and reduces the communication cost by 20%~90% compared with the existing state-of-the-art works.http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025115/privacy-preservinglarge language modelsecret sharingsecure multi-party computationcryptographic inference
spellingShingle CHENG Ke
XIA Yuheng
DAI Chuanyun
FU Jiaxuan
ZHU Xinghui
SHEN Yulong
Cryptographic inference for large language model via secret sharing
Tongxin xuebao
privacy-preserving
large language model
secret sharing
secure multi-party computation
cryptographic inference
title Cryptographic inference for large language model via secret sharing
title_full Cryptographic inference for large language model via secret sharing
title_fullStr Cryptographic inference for large language model via secret sharing
title_full_unstemmed Cryptographic inference for large language model via secret sharing
title_short Cryptographic inference for large language model via secret sharing
title_sort cryptographic inference for large language model via secret sharing
topic privacy-preserving
large language model
secret sharing
secure multi-party computation
cryptographic inference
url http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025115/
work_keys_str_mv AT chengke cryptographicinferenceforlargelanguagemodelviasecretsharing
AT xiayuheng cryptographicinferenceforlargelanguagemodelviasecretsharing
AT daichuanyun cryptographicinferenceforlargelanguagemodelviasecretsharing
AT fujiaxuan cryptographicinferenceforlargelanguagemodelviasecretsharing
AT zhuxinghui cryptographicinferenceforlargelanguagemodelviasecretsharing
AT shenyulong cryptographicinferenceforlargelanguagemodelviasecretsharing