---
格式版本: 2
标题: "MAGMA: Mixture-Model Adaptive Gaussian Model Acceleration"
原文链接: "https://arxiv.org/abs/2608.18366"
发布日期: "2026-08-18"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Tue, 18 Aug 2026 22:37:36 UTC (1,333 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T14:35:55+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T14:32:56+08:00"
入库时间: "2026-08-20T06:35:55.537Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=AMD&searchtype=all"
匹配关键词:
  []
相关厂家:
  - "AMD"
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 8
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "该资料为arXiv学术论文，主题是FPGA高斯混合模型加速，与超节点/AI Rack/机柜级AI基础设施等核心关注点完全无关，仅提及AMD FPGA芯片，无任何相关技术或商业信号。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T14:36:21+08:00"
AI主题相关性: 0
AI来源权威性: 6
AI新颖性: 2
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 0
AI摘要: "MAGMA 提出一种基于 FPGA 的混合高斯模型加速架构，可从流式 RGB 像素中同时完成推理与在线 EM 参数自适应。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:21:27.935Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18366"
---

## Computer Science > Hardware Architecture

## Title:MAGMA: Mixture-Model Adaptive Gaussian Model Acceleration

[View PDF](https://arxiv.org/pdf/2608.18366) [HTML (experimental)](https://arxiv.org/html/2608.18366v1)

> Abstract:Conventional FPGA-based Gaussian Mixture Model (GMM) accelerators use offline-trained, fixed parameters, limiting their ability to adapt to evolving scene statistics in long-lived edge systems. We present MAGMA, a fully synthesizable fixed-point FPGA architecture that performs concurrent GMM inference and online Expectation-Maximization (EM) parameter adaptation from a streaming RGB pixel input. MAGMA combines a pipelined inference datapath with a background update engine using hardware-friendly transcendental approximations---a range-reduced Chebyshev exponential, a CLZ-based logarithm, and a shift-and-subtract divider---alongside guards against variance collapse and cluster death that stabilize online fixed-point EM. Implemented on an AMD Spartan-7 XC7S50 with $K=4$ clusters, MAGMA runs at 74.49~MHz using 7,779 LUTs, 91 DSPs, and no block RAM, consuming 274~mW. It achieves an $11.8\times$ inference speedup and an $81\times$ M-step speedup over software, while spatial subsampling reduces per-update pixel volume by $40\times$ with minimal impact on EM convergence. Under a synthetic non-stationary scene, MAGMA's online adaptation improves mean pixel accuracy over a static baseline (81.5\\% vs.\\ 79.7\\%), demonstrating that full online GMM learning is achievable on a commodity edge FPGA.

| Subjects: | Hardware Architecture (cs.AR) |
| --- | --- |
| Cite as: | [arXiv:2608.18366](https://arxiv.org/abs/2608.18366) \[cs.AR\] |
|  | (or [arXiv:2608.18366v1](https://arxiv.org/abs/2608.18366v1) \[cs.AR\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18366](https://doi.org/10.48550/arXiv.2608.18366) |

## Submission history

From: Peter Forcha Mr \[[view email](https://arxiv.org/show-email/b50d3593/2608.18366)\]  
**\[v1\]** Tue, 18 Aug 2026 22:37:36 UTC (1,333 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18366) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))