LightSplit : Practical Privacy-Preserving Split Learning via Orthogonal Projections

Mert Cihangiroglu 1 , Alessandro Pegoraro 2 , Phillip Rieger 2 , Antonino Nocera 1 , Ahmad-Reza Sadeghi 2 · 2026

Read LightSplit : Practical Privacy-Preserving Split Learning via Orthogonal Projections by Mert Cihangiroglu 1 , Alessandro Pegoraro 2 , Phillip Rieger 2 , Antonino Nocera 1 , Ahmad-Reza Sadeghi 2 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-13265.txt.

Introduction

Mert Cihangiroglu and colleagues present LightSplit, a privacy guard for split learning between clients and a server. Raw activations at the split point cost bandwidth and leak data to reconstruction attacks. A fixed orthogonal random projection squeezes activations before sending, acting as a bottleneck that drops sample specific detail. At up to 32 fold compression it keeps at least 94 percent accuracy on several image and tabular tasks. The strongest co trained attack loses most of its reconstruction power.

Catalog record

CREATORMert Cihangiroglu 1 , Alessandro Pegoraro 2 , Phillip Rieger 2 , Antonino Nocera 1 , Ahmad-Reza Sadeghi 2
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE131504
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.13265

Browse the full library · About the Tower of Babel