The Transformer Revolution, Part 1: Dynamic Processing through Output- Weight Interconnections

📄 arXiv: 2608.03921v1 📥 PDF

作者: Marco Giunti, Fabrizia Giulia Garavaglia

分类: cs.AI, cs.NE

发布日期: 2026-08-04

备注: v1: 7 Sections, References, Appendix, Tables (2 tables), Figures Part A (6 figures), Figures Part B (42 figures), Figures Part C (5 figures)


💡 一句话要点

提出SIDPP以解决Transformer推理中的动态处理问题

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: Transformer 动态处理 SIDPP 输出权重互连 自然语言处理 模型解释性 人类语言处理

📋 核心要点

  1. 现有方法往往将大型语言模型视为简单的统计重现,缺乏对其动态处理能力的深入理解。
  2. 论文提出SIDPP概念,强调Transformer在推理时通过动态生成的变换参数进行交互式处理。
  3. 研究表明,动态处理的贡献随着提示长度增加而显著提升,可能超越静态处理的效果。

📝 摘要(中文)

本文对Transformer在推理过程中的新解释提出了SIDPP(序列级交互动态并行处理)概念。与将大型语言模型视为简单重现统计规律的观点相对,作者认为Transformer在推理时构建并应用依赖于提示的变换,这些变换的参数在推理过程中生成。通过输出权重互连,某些网络的输出决定其他网络的权重,从而实现动态处理。动态处理的贡献随着提示长度的增加而增长,可能等于或超过静态处理。最后,作者推测人类神经系统可能具备实现SIDPP的机制,暗示人类语言处理可能与Transformer的功能架构相似。

🔬 方法详解

问题定义:本文旨在解决对Transformer推理过程的传统理解不足,尤其是其动态处理能力的低估。现有方法往往忽视了模型在推理时生成的动态变换参数。

核心思路:作者提出SIDPP概念,认为Transformer通过提示生成的动态变换参数进行交互式处理,从而实现更复杂的概念转化。这样的设计使得模型能够根据输入序列灵活调整其处理方式。

技术框架:Transformer的架构中引入了输出权重互连机制,允许某些网络的输出影响其他网络的权重。整体流程包括输入序列的处理、动态变换参数的生成以及基于这些参数的概念转化。

关键创新:最重要的创新在于输出权重互连的引入,使得模型能够在推理时动态调整其处理方式,与传统的静态处理方法形成鲜明对比。

关键设计:论文中涉及的关键设计包括动态生成的变换参数、网络结构的输出权重互连机制,以及在不同提示长度下的动态处理能力评估。具体的参数设置和损失函数设计尚未详细披露。

🖼️ 关键图片

img_0
img_1
img_2

📊 实验亮点

实验结果表明,随着提示长度的增加,动态处理的贡献显著提升,甚至可能超过静态处理。具体性能数据尚未提供,但作者强调了强提示敏感性现象的存在,表明模型在处理复杂任务时的优势。

🎯 应用场景

该研究的潜在应用领域包括自然语言处理、对话系统和智能助手等。通过实现更灵活的动态处理,Transformer模型可以在更复杂的语言任务中表现出更高的准确性和适应性,未来可能推动更小型且可持续的系统设计。

📄 摘要(原文)

This paper offers a new interpretation of the Transformer during inference. Against the "stochastic parrot" view that large language models merely reproduce statistical regularities learned in training, we argue that Transformers construct and apply prompt-dependent transformations whose parameters are generated during inference. We call this form of computation SIDPP: Sequence-level Interactive Dynamic Parallel Processing. The Transformer is interpreted as a system that transforms concepts by means of concepts. Token vectors are the concepts to be transformed; parameterized transformations defined by matrices and vectors are the transforming concepts. These may be static, when fixed through training, or dynamic, when generated from the input sequence. Mechanically, they correspond to groups of simple neural networks. The Transformer's architectural novelty lies in output-weight interconnections, through which the outputs of some networks determine the weights of others, alongside ordinary output-input interconnections. By means of these interconnections, the system constructs transformations from the prompt and uses them to modify token representations. The contribution of dynamic processing grows with prompt length and may equal or exceed that of static processing, a phenomenon we call strong prompt sensitivity. This account bears on interpretability, predictability, control, and the design of smaller, more sustainable systems. Finally, since the human neural system possesses the mechanisms required to implement SIDPP, we argue that a form of SIDPP may, in principle, be neurally realized in the cerebral cortex. We therefore conjecture that human language processing may itself be a form of SIDPP produced by a functional architecture relevantly similar to that of the Transformer.