|
所在平台: Udemy |
课程主页: https://www.udemy.com/course/federated_learning/
课程评论：没有评论
**课程名称：** 联邦学习 (Federated Learning) **课程概述：** 本课程将从神经网络（NN）的基本概念及其工作原理入手，并使用PyTorch从零开始实现神经网络。随后，将简要介绍联邦学习（FL）的架构。接着，课程将指导您在客户端设备上加载数据集，分别处理IID、Non-IID以及Non-IID但数据量不均衡的场景。课程还将提供一个PySyft的简要教程，演示如何在客户端和服务器之间发送和接收模型与数据集。 本课程将通过研读原始论文中的技术和算法，并逐行实现，来深入教授联邦学习。具体而言，我们将实现FedAvg、FedSGD、FedProx和FedDANE算法。您将学习差分隐私（DP）及其在联邦学习中的应用，并实现带有DP的FedAvg。 通过本课程，您将学会如何在本地和云端实现联邦学习技术。在云端环境下，我们将使用Google Cloud Platform（GCP）来创建和配置实验所需的各种实例。 **学习成果：** 完成本课程后，您将能够： * 实现不同的联邦学习技术。 * 构建自己的优化器和联邦学习技术。 * 在本地和云端运行联邦学习实验。 * 理解神经网络基础及实现。 * 掌握使用PySyft进行模型和数据在客户端与服务器间的传输。 * 学习并实现差分隐私在联邦学习中的应用。
The course starts by introducing you to the main concepts in Neural Networks (NN) and how do they work. Then we will implement a NN from scratch using Pytorch. After that, a quick introduction to Federated Learning architecture. Then, we will start by loading the dataset on the devices in IID, non-IID, and non-IID and unbalanced settings followed by a quick tutorial on PySyft to show you how to send and receive the models and the datasets between the clients and the server.This course will teach you Federated Learning (FL) by looking at the original papers' techniques and algorithms then implement them line by line. In particular, we will implement FedAvg, FedSGD, FedProx, and FedDANE. You will learn about Differential Privacy (DP) and how to add it to FL, then we will implement FedAvg using DP. In this course, you will learn how to implement FL techniques locally and on the cloud. For the cloud setting, we will use Google Cloud Platform to create and configure all the instances that we will use in our experiments. By the end of this course, you will be able to implement different FL techniques and even build your own optimizer and technique. You will be able to run your experiments locally and on the cloud.
