{"slug": "high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs", "title": "High School Sophomore Seeking arXiv Endorser for Vision Transformer MoE Paper (cs.LG / cs.CV)", "summary": "Hanyu (Olivia) Yu, a high school sophomore, is seeking an arXiv endorser in cs.LG or cs.CV for her independent preprint on Vision Transformer Mixture of Experts (MoE) routing. Her paper proposes using three scalar signals from the self-attention matrix as a routing prior that anneals to zero during training, finding that routing diversity at initialization prevents early collapse, while semantic attention shapes post-training specialization. The work includes 3-seed experiments, four ablation conditions, and reports a non-significant p > 0.05 result.", "body_md": "Hi everyone,\n\nI am a high school sophomore looking for an arXiv endorser in **cs.LG** (Machine Learning) or **cs.CV** (Computer Vision) to help upload my independent preprint.\n\n**Brief Overview:**\n\nStandard MoE routers discard the self-attention matrix before routing, despite its existing structural information on patch semantics. My paper introduces three scalar signals from this matrix (entropy, cross-head variance, and CLS-similarity) as a routing prior that anneals to zero during training. Through random prior ablations and signal dropout experiments, I found that routing diversity at initialization—rather than semantic content—is what drives early collapse prevention. Semantic attention primarily shapes post-training expert specialization. The work includes 3-seed experiments, four ablation conditions, and honest reporting of a non-significant p > 0.05 result.\n\n**Endorsement Criteria:**\n\nTo endorse this submission, arXiv requires you to have 3+ papers submitted to `cs.*`\n\ncategories within the last 5 years (with at least one paper being older than 3 months).\n\nIf you are willing to endorse or have any feedback, please reply here or reach out so I can share the arXiv submission link. Thank you so much for your time and consideration!\n\nBest regards,\n\nHanyu (Olivia) Yu", "url": "https://wpnews.pro/news/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs", "canonical_source": "https://discuss.huggingface.co/t/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs-lg-cs-cv/178377#post_1", "published_at": "2026-08-01 21:35:48+00:00", "updated_at": "2026-08-01 22:02:53.994732+00:00", "lang": "en", "topics": ["machine-learning", "computer-vision", "ai-research"], "entities": ["Hanyu (Olivia) Yu", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs", "markdown": "https://wpnews.pro/news/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs.md", "text": "https://wpnews.pro/news/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs.txt", "jsonld": "https://wpnews.pro/news/high-school-sophomore-seeking-arxiv-endorser-for-vision-transformer-moe-paper-cs.jsonld"}}