Federated Learning Experiment | 李大饼的博客

发现东亚书摘缔造和平1919书摘 2019阅读书单 2016年阅读书单

2016年发出明信片记录 2016年收到明信片记录

一个豆瓣爬虫+社交分析 warmshowers利用account实现账户管理

Federated Learning Experiment Federated Learning Experiment Federated Learning Experiment Federated Learning Secure Agre'ga'ti Actively Secure OT Extension with Optimal Overhead Covert Security with Public Verifiability Fast Cut-and-Choose Based Protocols 论文十问 Improve Garbled Circuit:Free XOR Gates and Applications Efficient Garbling from a Fixed-Key Blockcipher Secure Two-Party Computation Is Practical Secure Multi-Party Computation Survey More Efficient Oblivious Transfer and Extensions for Faster Secure Computation 研究生求生手册

广州攻略日本骑行攻略日本骑行攻略

有所不为的反叛者

PC 2

2016年发出明信片记录 2016年收到明信片记录

warmshwoers 1

warmshowers利用account实现账户管理

android 1

warmshowers利用account实现账户管理

学习 1

warmshowers利用account实现账户管理

笔记 13

有所不为的反叛者 Federated Learning Experiment Federated Learning Experiment Federated Learning Experiment Federated Learning Covert Security with Public Verifiability Fast Cut-and-Choose Based Protocols Improve Garbled Circuit:Free XOR Gates and Applications Efficient Garbling from a Fixed-Key Blockcipher Secure Two-Party Computation Is Practical Secure Multi-Party Computation Survey More Efficient Oblivious Transfer and Extensions for Faster Secure Computation 研究生求生手册

心得 3

Secure Agre'ga'ti 论文十问研究生求生手册

书单 3

发现东亚书摘缔造和平1919书摘 2019阅读书单

2019 3

发现东亚书摘缔造和平1919书摘 2019阅读书单

阅读 3

发现东亚书摘缔造和平1919书摘 2019阅读书单

论文 14

Federated Learning Experiment Federated Learning Experiment Federated Learning Experiment Federated Learning Secure Agre'ga'ti Actively Secure OT Extension with Optimal Overhead Covert Security with Public Verifiability Fast Cut-and-Choose Based Protocols 论文十问 Improve Garbled Circuit:Free XOR Gates and Applications Efficient Garbling from a Fixed-Key Blockcipher Secure Two-Party Computation Is Practical Secure Multi-Party Computation Survey More Efficient Oblivious Transfer and Extensions for Faster Secure Computation

安全多方计算 8

Actively Secure OT Extension with Optimal Overhead Covert Security with Public Verifiability Fast Cut-and-Choose Based Protocols Improve Garbled Circuit:Free XOR Gates and Applications Efficient Garbling from a Fixed-Key Blockcipher Secure Two-Party Computation Is Practical Secure Multi-Party Computation Survey More Efficient Oblivious Transfer and Extensions for Faster Secure Computation

garbled-circuit 1

More Efficient Oblivious Transfer and Extensions for Faster Secure Computation

密码学 8

Actively Secure OT Extension with Optimal Overhead Covert Security with Public Verifiability Fast Cut-and-Choose Based Protocols Improve Garbled Circuit:Free XOR Gates and Applications Efficient Garbling from a Fixed-Key Blockcipher Secure Two-Party Computation Is Practical Secure Multi-Party Computation Survey More Efficient Oblivious Transfer and Extensions for Faster Secure Computation

爬虫 1

一个豆瓣爬虫+社交分析

scrapy 1

一个豆瓣爬虫+社交分析

豆瓣 1

一个豆瓣爬虫+社交分析

Python 1

一个豆瓣爬虫+社交分析

巴尔干 1

日本骑行攻略

欧洲 1

日本骑行攻略

骑行 2

日本骑行攻略日本骑行攻略

攻略 3

广州攻略日本骑行攻略日本骑行攻略

日本 1

日本骑行攻略

亚洲 1

日本骑行攻略

OT 1

Actively Secure OT Extension with Optimal Overhead

国内 1

广州攻略

广东 1

广州攻略

FederatedLearning 3

Federated Learning Experiment Federated Learning Experiment Federated Learning

联邦学习 3

Federated Learning Experiment Federated Learning Experiment Federated Learning

Meachine 1

Federated Learning Experiment

learning 1

Federated Learning Experiment

历史 1

有所不为的反叛者

Federated Learning Experiment

2019年07月22日

Lecture 2 Where does the error come from?

Bias and Variance of Esstimator

样本的无偏估计错误来自于 bias（期望值和真实值有多接近）和 variance（点的分布有多开）

通过bias和variance可以判断是否underfitting还是overfitting。 bias大，需要重新设计model，增减更多的feature，设计更负载的模型。variance大的话需要增加训练的书籍量

Lecture 3 Gradient Descent

如何把gradient descent做得更好

Adaptive Descent

Adagrad

$w^{t+1} \leftarrow w^{t}-\frac{\eta^{t}}{\sigma^{t}} g^{t}$ $\sigma^{t} 是过去所有微分值的root mean square$

e.g $\begin{array}{ll}{w^{1} \leftarrow w^{0}-\frac{\eta^{0}}{\sigma^{0}} g^{0}} & {\sigma^{0}=\sqrt{\left(g^{0}\right)^{2}}} \\ {w^{2} \leftarrow w^{1}-\frac{\eta^{1}}{\sigma^{1}} g^{1}} & {\sigma^{1}=\sqrt{\frac{1}{2}\left[\left(g^{0}\right)^{2}+\left(g^{1}\right)^{2}\right]}}\end{array}$

如何解释这个方法：造成一个反差效果

Stochastic Gradient Descent

选一个点来计算loss来进行gradient Descent

Feature Scaling

调整参数分布，让分布相同

Lecture 4 classification

数据构造

MNIST测试集中一共有60000张图片 0：5923 1：6742 2：5958 3：6131 4：5842 5：5421 6：5918 7：6265 8：5851 9：5949