English(EN) SG-Blend: Learning an Interpolation Between Improved Swish and GELU for Robust Neural Representations

新的SG-Blend激活函数提高了神经网络的鲁棒性

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-11 04:00

研究人员推出了一种新颖的自适应激活函数SG-Blend，旨在改善神经网络表示。SG-Blend在Swish和GELU激活函数之间进行插值，允许每一层学习其在该连续体上的最佳位置。这种方法在BERT风格的IMDB分类任务上显示出较低的方差并保持了峰值准确率，同时在WikiText103的自回归预训练中也实现了较低的验证困惑度。该方法的有效性超出了自然语言处理的范畴，在计算机视觉和其他领域也显示出鲁棒的泛化能力。 AI

影响 SG-Blend的自适应性和已证实的鲁棒性有望在各种AI领域实现更高效、更可靠的神经网络训练。

排序理由该集群包含一篇详细介绍神经网络新激活函数的论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 13 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍神经网络新激活函数的论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Gaurav Sarkar, Syed Affan Daimi, Jay Gala, Subarna Tripathi · 2026-09-11 04:00

SG-Blend：学习改进版Swish与GELU之间的插值以获得鲁棒的神经表示

arXiv:2505.23942v2 Announce Type: replace Abstract: Prevailing activation functions such as Swish and GELU tend toward domain-specific optima, Swish was discovered via neural architecture search on vision benchmarks, while GELU dominates transformer-based language models, and nei…

本文如何被排名

报道来源 [1]

SG-Blend：学习改进版Swish与GELU之间的插值以获得鲁棒的神经表示

相关实体

相关话题