内容库 / 论文

arXiv:2608.00574 · 多模态

Relax Within, Balance Across: Geometry-Guided Load Balancing for Vision-Language Mixture-of-Experts

内部松弛、跨模态均衡：面向视觉-语言 Mixture-of-Experts 的几何引导负载均衡

Relax Within, Balance Across: Geometry-Guided Load Balancing for Vision-Language Mixture-of-Experts

类型：arxiv
标识：2608.00574
链接：https://arxiv.org/abs/2608.00574
主分类：multimodal
形态：method
被引：0
被引来源：Semantic Scholar
S2被引：0
影响力被引：0
TLDR：Across four split backbones, ReBA lowers load on every reported benchmark input while keeping mean task accuracy comparable to Std-Aux, and lowers average load over the tested range and worst physical load under resolution and tiling shifts.
待LLM分类：否
标题中文：内部松弛、跨模态均衡：面向视觉-语言 Mixture-of-Experts 的几何引导负载均衡
TLDR中文：在四种拆分 backbone 上，ReBA 降低所有报告 benchmark 输入的负载，同时保持与 Std-Aux 相当的平均任务准确率，并在分辨率与分块变化下降低测试范围内的平均负载与最差物理负载
来源文件：
/inbox/tom/_candidates/2026-08-04-agent-rag-longcontext-candidates.json
[S2 enrich]

相关 · 按语义匹配

笔记精读与批判：SenseNova-Vision — Vision as Unified Multimodal Generation

笔记 VaLR: Vision-aligned Latent Reasoning for Multi-modal LLM

仓库 EvolvingLMMs-Lab/lmms-eval

笔记精读与批判：OPPO — Oriented Pickup Preference Optimization for Multimodal Hallucination Mitigation

笔记 LoomVideo · Unifying Multimodal Inputs into Video Generation and Editing · 批判性精读

文档 Tom 评 flyP 的产出：TimeLens2 + ShotPlan 2 件连读精读

arXiv

2608.00574

TLDR

在四种拆分 backbone 上，ReBA 降低所有报告 benchmark 输入的负载，同时保持与 Std-Aux 相当的平均任务准确率，并在分辨率与分块变化下降低测试范围内的平均负载与最差物理负载Across four split backbones, ReBA lowers load on every reported benchmark input while keeping mean task accuracy comparable to Std-Aux, and lowers average load over the tested range and worst physical load under resolution and tiling shifts.

被引 · S2

0

S2 0

影响力 0

相关解读

arXiv ↗ 中文全文