跳到正文
arXiv cs.AI· Mohamed Abdallah Salem·· 4 小时前AI 评分34

大语言模型能否优化预训练神经网络权重?基于 YOLO26 目标检测的无梯度权重编辑多阶段实证评估

Can Large Language Models Optimize Pretrained Neural Network Weights? A Multi-Phase Empirical Evaluation of Gradient-Free Editing in YOLO26 Object Detection

AI 导读

一项多阶段实证研究以 YOLO26 模型和 MS COCO 数据集评估了确定性缩放、差分进化、CMA-ES 及 LLM 引导等无梯度权重编辑方法。

正文 · 原文

View PDF HTML (experimental)

Abstract:Direct weight editing offers a gradient-free alternative to fine-tuning pretrained detectors, but its generalization and the value of large language model (LLM) guidance remain uncertain. We evaluate five experimental phases using YOLO26 models and MS COCO, comparing deterministic scaling, random perturbations, differential evolution, CMA-ES, structured interventions, real local-LLM proposals, failure-aware guidance, and hybrid LLM-CMA-ES against a matched random-policy control. Optimization uses native Ultralytics mAP50-95 on 1,000 COCO Val2017 images; selected candidates are assessed on a disjoint, previously examined 1,500-image subset, with COCO API AP50-95 reported separately where available. Increasing the CMA-ES budget from 8 to 250 evaluations raised development gains from +0.11 to +0.28 percentage points, while selection changes remained negative (-0.03 to -0.12 points); the 250-evaluation result uses one seed. The hybrid LLM approach showed no statistically established advantage over random-policy guidance (p = 0.612, 0.369, and 0.694 at budgets 8, 25, and 50). Five-seed YOLO26m replication yielded a COCO API AP difference of +0.004 points (95% CI: -0.026 to +0.034). Of 225 archived LLM proposals, 53 satisfy the final strict schema; early tensor-level execution cannot be fully verified. Genuine LLM-mediated weight editing is feasible, but the tested approaches did not demonstrate consistent transfer beyond the development subset or a measurable LLM advantage.
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.17789 [cs.CV]
  (or arXiv:2611.17789v1 [cs.CV] for this version)
  https://doi.org/10.48550/arXiv.2611.17789

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Mohamed Abdallah Salem [view email]
[v1] Fri, 13 Nov 2026 06:05:11 UTC (245 KB)

来源:arXiv cs.AI · arxiv.org