2024.1.14周报_pinn + resnet-CSDN博客

本周我阅读了一篇题目为Deep Residual Learning for Image Recognition的文献，文章的贡献是作者提出了残差网络的思想，且证明了更深层的残差网络具有比VGG网络更低的复杂度和更高的准确性，同时，残差网络实现了更容易的训练过程。其次，对PINN进行了初步的认识，PINN 是一种科学机器在传统数值领域的应用方法，特别是用于解决与偏微分方程相关的各种问题，包括方程求解、参数反演、模型发现、控制与优化等。

This week, I read a paper titled "Deep Residual Learning for Image Recognition." The contribution of the paper is that the author introduces the concept of residual networks and demonstrates that deeper residual networks have lower complexity and higher accuracy compared to VGG networks. Additionally, residual networks achieve a more straightforward training process. Furthermore, I gained preliminary insights into Physics-Informed Neural Networks (PINN). PINN is an application of scientific machine learning in traditional numerical domains, particularly for solving various problems related to partial differential equations. This includes equation solving, parameter inversion, model discovery, control, and optimization.

一、文献阅读

1、题目

题目：Deep Residual Learning for Image Recognition

链接：https://arxiv.org/abs/1512.03385

2、摘要

本文展示了一种残差学习框架，能够简化使那些非常深的网络的训练，该框架使得层能根据其输入来学习残差函数而非原始函数。作者提出了残差网络的思想，且证明了更深层的残差网络具有比VGG网络更低的复杂度和更高的准确性。同时，残差网络实现了更容易的训练过程。

This article presents a residual learning framework that simplifies the training of very deep networks. The framework allows layers to learn residual functions based on their inputs rather than the original functions. The author introduces the concept of residual networks and demonstrates that deeper residual networks have lower complexity and higher accuracy compared to VGG networks. Additionally, residual networks achieve a more straightforward training process.