---
id: 20260620-T0-08
title: "ITNet：一种统一卷积、注意力与循环机制的新网络架构"
title_en: "ITNet: A Unified Architecture Subsuming Convolution, Attention, and Recurrence"
url: https://ai.daily.yangsir.net/daily/20260620-T0-08
issue_date: 2026-06-20
publish_date: 2026-06-19T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.19538
---

# ITNet：一种统一卷积、注意力与循环机制的新网络架构

新研究提出了ITNet，这是一种可学习的积分变换网络。该架构在数学上统一了卷积网络（局部性）、循环网络（序列记忆）和Transformer（成对交互）的特性。实验表明，ITNet在长序列建模任务中表现优异，且能有效整合不同归纳偏置。这为设计下一代通用神经网络基础架构提供了理论基础。

## English Version

**ITNet: A Unified Architecture Subsuming Convolution, Attention, and Recurrence**

New research introduces ITNet, a learnable integral transform network. It mathematically unifies the inductive biases of Convolutional networks (locality), Recurrent networks (sequential memory), and Transformers (pairwise interaction). Experiments show ITNet excels in long-sequence modeling and effectively integrates diverse biases. This provides a theoretical foundation for next-generation universal neural network architectures.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.19538)

**详情页**：https://ai.daily.yangsir.net/daily/20260620-T0-08

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*