R07: Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware

Status: tracked
Year: 2019
Authors: Florian Tramèr, Dan Boneh
External source: Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware

Source contribution

Slalom outsources masked linear work and uses local verification within a trusted execution partition.

PLLM perspective

Reproduce the original blinding and verification boundary. Register PLLM trusted-client/dealer adaptation separately; port modular kernels and verification state without changing the numerical graph silently.

The full Slalom paper is acquired and digest-locked. Original PLLM manuscript runs remain an adaptation, not a fresh reproduction of Slalom; no upstream artifact is acquired or executed.

This paper remains tracked. Named modules below are implementation targets, not finished PLLM components; no reproduction, integration, assurance, or matched benchmark is claimed.

The source's boundary remains binding: A trusted client is not a reproduced TEE deployment. Static non-collusion, integrity and later compromise require distinct claims.

PLLM modules and documentation

Target module identifiers: protocol.masked_linear, verify.preprocessed_linear. These identifiers describe scoped work unless this page explicitly links an implementation source. Dependencies: None.

• Research backlog
• Reimplementation backlog
• Clean-room workflow

Backlog

This source is assigned to reimplementation priority 1, Delegated matrix-vector execution. Its source-registry order is 7. Follow current gates and unfinished work in the research backlog.