技术

FlowGuard

更新于 2026/7/28 17:26:51

简介

A lightweight inference-time framework for detecting harmful inputs in multimodal large language models by monitoring internal cross-modal consistency using FlowVectors derived from Partial Information Decomposition.

别名FlowGuard framework

实体关系

REFERENCESFlowGuard → cross-modal consistency可信度 98%

FlowGuard detects harmful inputs by monitoring internal multimodal consistency

REFERENCESFlowGuard → Partial Information Decomposition可信度 95%

FlowGuard derives FlowVectors inspired by Partial Information Decomposition

COOPERATES_WITHFlowGuard → FlowVectors可信度 92%

FlowGuard derives FlowVectors... that quantify cross-modal redundancy, synergy, and modality-specific dominance

UPSTREAMFlowGuard → Multimodal large language models可信度 90%

FlowGuard is a framework designed to secure multimodal large language models

UPSTREAMarXiv → FlowGuard可信度 85%

The FlowGuard paper is published on arXiv (arXiv:2607.21600v1)

关联事件

来源

暂无来源链接。