1.联邦学习相比于传统分布式学习有哪些区别？
- 数据保留在用户本地，用户对他们的设备和数据有绝对的控制权
- 设备不稳定，如设备的计算能力有强弱之分，设备存在意外掉线的情况
- 通信代价远大于计算代价
- 设备的数据并非独立同分布，绝大部分为Non-IID数据
- 节点负载不平衡，如一个用户拥有上千张照片，而另一个用户只有几十张照片
2.Non-IID（非独立同分布）数据
Non-identical（非同分布）：由于设备归属的用户、场景各不相同，导致数据分布往往差异较大。
Non-independent（非独立）：由于用户群体、地域等因素使数据间存在关联，导致非独立性。
参考链接：
3.Federated Averaging
FedAvg算法就目前的联邦学习而言应用较为广泛。相比于并行梯度下降算法FedSGD，FedAvg通过增加本地计算减少了通信次数，但FedAvg收敛速度要慢于FedSGD
值得一提的是，FedAvg目前被证明可以用在Non-IID数据集上，这对联邦学习而言是非常重要的。
参考链接：
4.Proof of Work与Proof of Stake
Proof of Work
PoW（工作量证明）：要求用户进行耗费一定时间的复杂运算，并且答案能够被服务方快速验算，以此耗用的时间、设备与能源作为担保成本，确保服务与资源是被真正的需求所使用。
缺点：
- 非常耗电，浪费能源
- 区块确认时间难以缩短
- 新的区块链必须找到一种不同的散列算法，否则就会面临比特币的算力攻击
- 容易产生分叉，需要等待多个确认
- 永远没有最终性，需要检查点机制来弥补最终性
- 如今比特币网络中出现算力相对集中的矿池，比特币的去中心化受到威胁
Proof of Stake
PoS（权益证明）：权益证明并不要求用户拥有一定的算力（Work），而要求拥有一定的财产（Stake)，这也就意味着用户如果作弊损害了系统，则我的财产会相应的贬值或扣除部分财产，这变相的让我付出了代价。
缺点：无利益攻击（nothing-at-the-stake attack）
对于PoS共识机制的货币，你越有钱，作恶付出的代价就越大；然而对于没有钱的人，作恶代价很小，所以一些作恶行为对他们而言是有益的
5.联邦学习正面临什么问题？（研究方向有哪些）
Non-IID背景下通信代价过高
在联邦学习中，计算代价远没有通信代价高。FedAvg正是利用了更多的本地计算来减少通信次数，因此其效率也更好于FedSGD
隐私保护
无论是FedAvg还是FedSGD都需要将模型的参数上传到服务器或聚合节点，看似用户数据都保留在本地，但FedAvg上传的权值或FedSGD上传的梯度都是数据在多个函数的映射，存在模型参数反推用户数据的可能
解决方案：差分隐私（容易造成模型精度下降）
鲁棒性
拜占庭错误：存在恶意用户向模型注入错误的参数或数据，导致模型整体精度下降等影响
非拜占庭错误：非恶意用户的设备出现故障，但并没有停止上传参数，导致全局模型掺杂了不良数据
抵御方式：校验集检测（用户本地数据与校验集数据统计分布不同，尽管本地训练精度很高但校验精度可能会很低，检测存在缺陷，除非将所有权值聚合再校验）、梯度检测（前提是IID数据，否则梯度偏移可能很大）
持续更新中…
Can I simply say what a comfort to find someone that actually understands what theyre discussing on the net. You actually realize how to bring a problem to light and make it important. More and more people really need to read this and understand this side of your story. I was surprised that youre not more popular since you definitely possess the gift.
Great info and right to the point. I am not sure if this is actually the best place to ask but do you people have any ideea where to employ some professional writers? Thank you 🙂
I think this is one of the such a lot vital information for me. And i’m glad reading your article. However want to statement on some basic issues, The web site style is perfect, the articles is really excellent : D. Just right process, cheers
Great write-up, I am regular visitor of one’s web site, maintain up the nice operate, and It’s going to be a regular visitor for a long time.
Rattling nice design and superb subject matter, very little else we need : D.
good post.Never knew this, thanks for letting me know.
My husband and i ended up being very glad that Emmanuel managed to round up his analysis through your ideas he grabbed using your site. It is now and again perplexing to simply be handing out guidelines that a number of people might have been making money from. And we also fully grasp we now have the blog owner to appreciate for this. The illustrations you have made, the simple web site navigation, the friendships your site help to promote – it is most astounding, and it’s making our son and us imagine that this matter is pleasurable, which is especially vital. Many thanks for the whole thing!
Attractive section of content. I just stumbled upon your site and in accession capital to assert that I acquire in fact enjoyed account your blog posts. Anyway I will be subscribing to your augment and even I achievement you access consistently rapidly.
Hello I am so thrilled I found your site, I really found you by error, while I was searching on Digg for something else, Regardless I am here now and would just like to say thanks for a incredible post and a all round enjoyable blog (I also love the theme/design), I don’t have time to look over it all at the moment but I have book-marked it and also included your RSS feeds, so when I have time I will be back to read a lot more, Please do keep up the great work.
We are a group of volunteers and opening a new scheme in our community. Your website provided us with valuable info to work on. You’ve done a formidable job and our entire community will be grateful to you.