A few years ago, I secured a scholarship to pursue a PhD in Photonics modelling. Modelling and optimization of optical/photonics waveguides are very important for many upcoming applications, like optical waveguides used for guiding light in Augmented Reality applications. Seeing such cool applications motivated me a lot to start my PhD journey.

几年前，我获得了奖学金，以攻读光子学建模博士学位。 光学/光子波导的建模和优化对于许多即将到来的应用非常重要，例如在增强现实应用中用于引导光的光波导。 看到如此出色的应用程序激励了我很多，从而开始了我的博士学位之旅。

Like any other PhD researcher, I went to one place where everyone goes when they are not sure what they are looking for and how to find what they are looking for. Yes, you guessed it right — GOOGLE!!

像任何其他博士研究人员一样，我去了一个地方，每个人都不确定自己在寻找什么以及如何找到他们想要的东西。 是的，您猜对了-GOOGLE！

Luckily for me, I already knew that I was looking for some sort of open-source codes/libraries or commercial software to get my hands dirty with photonics modelling applications.

对我来说幸运的是，我已经知道我正在寻找某种开源代码/库或商业软件，以使我的手与光子建模应用程序隔离。

令我惊讶的是，我发现了…… (To my surprise what I found…)

• Commercial softwares- Comsol/Lumerical were expensive for me to buy at that time, and I was not sure to even ask my supervisor to buy as I was not sure which one I will need.

  商业软件 -当时，我购买的Comsol / Lumerical 软件价格昂贵，而且我不确定要购买哪一款，因此我什至不确定要问主管。

• Open-source libraries- There are some open-source libraries like Meep, but I think we need to have more. Google, Facebook, Microsoft all adopting this new trend of making their technologies open source for all to learn. But are the big companies in the optical modelling domain ready to go open-source?

  开源库-有一些开源库，例如Meep，但我认为我们需要更多。 Google，Facebook，Microsoft都采用了这种新趋势，即将其技术开放给所有人学习。 但是，光学建模领域的大公司是否准备好开源？

• Fortran- I never bothered to learn about Fortran during my Bachelor’s, in fact, I can say that I chose not to hear or learn about Fortran as it was such an outdated language. The syntax is bad, no in-build plotting library, among other problems. Guess what? People still use it because it is fast and a lot of codes which are still being used were written in Fortran. But still Fortran…

  Fortran-在我的学士期间，我从不费心去了解Fortran，实际上，我可以说我选择不听或不了解Fortran，因为它是一种过时的语言。 语法不好，没有内置绘图库，还有其他问题。 原因是什么？ 人们仍然使用它，因为它速度很快，并且仍在使用的许多代码是用Fortran编写的。 但仍然是Fortran ...

• No optical engineers/scientists community- This hurted me the most. I couldn’t find any blog/website where optical engineer come and share their work and experiences regularly. Till date, I am looking for such websites sharing regular traffic where people talk about photonics (Let me know if you know any such place!).

  没有光学工程师/科学家社区 -这对我造成的伤害最大。 我找不到任何光学工程师来定期分享其工作和经验的博客/网站。 直到现在，我正在寻找这样的网站，这些网站共享经常的流量，人们谈论光子学(让我知道您是否知道这样的地方！)。

机器学习-流行语… (Machine Learning- the Buzz word…)

If you are reading this then there is a good chance that you have heard about today’s technologies buzz words-

如果您正在阅读本文，那么您很有可能听说过当今的技术流行语-

Artificial Intelligence, Machine Learning, Deep Learning

人工智能，机器学习，深度学习

Everyone is talking about it. Every new company is using it or wanting to use it. It doesn’t matter whether they need it or not, but they want to use it. Or atleast show that they are up to date with new technologies and making our life easier and better. I guess it is good marketing as well!!

每个人都在谈论它。 每个新公司都在使用它或想要使用它。 是否需要它并不重要，但他们想使用它。 或至少表明它们与新技术保持同步，使我们的生活更加轻松和美好。 我想这也是一个很好的市场！

Having heard these buzz words several times from Mark Zuckerberg, Bill Gates, Tim Cook, I thought there is no harm in looking for what exactly is this machine learning?

在听过马克·扎克伯格，比尔·盖茨，蒂姆·库克的这些热门词汇后，我认为寻找这种机器学习到底是什么没有害处？

What the hell is machine learning? Then I understood machine learning is actually statistics plus a lot of other things.

机器学习到底是什么？ 然后我了解到机器学习实际上是统计信息以及许多其他事情。

I started with free courses by Prof. Andrew Ng from Stanford and checking for some YouTube videos. I ended up spending more than £1000 doing courses from Udacity and Udemy. For a few months, these courses consumed my every weekend. Literally every weekend. I even had to say no to friends a few times to join for a party. Can you imagine!! I know it was stupid of me. Anyway, it is in the past now…

我先从斯坦福大学的吴安德教授提供免费课程，然后查看一些YouTube视频。 我最终花了1000多英镑从Udacity和Udemy上课。 几个月以来，这些课程每个周末都消耗了我。 从字面上看每个周末。 我什至不得不几次拒绝朋友参加聚会。 你可以想象！！ 我知道这很愚蠢。 无论如何，现在已经过去了……

机器学习+光子学… (Machine Learning + Photonics…)

First-year into my PhD, I was still looking for a topic which will become my PhD dissertation. Spent months reading various photonics research papers, understanding some initial optical modelling codes with as little help as possible and irritated with almost non-existent online optical scientist community.

进入博士学位的第一年，我仍在寻找一个将成为我的博士学位论文的主题。 花了几个月的时间阅读各种光子学研究论文，在几乎没有帮助的情况下理解了一些初始光学建模代码，并且对几乎不存在的在线光学科学家社区感到不快。

And that’s when it clicked me- Shall I try to use Machine Learning for Optical/Photonics Applications?

这就是单击它的时候-我是否应该尝试将机器学习用于光学/光子学应用程序？

Yeah, why not!! There is a big online community for machine learning and I had a background in optical engineering. I thought it might work. But remember I was still new in the Machine Learning field and barely scratched the surface to understand and use it efficiently. Then the questions start arising if I had to make it work-

是的，为什么不呢！ 有一个大型的机器学习在线社区，我有光学工程背景。 我认为这可能有效。 但是请记住，我仍然是机器学习领域的新手，几乎没有摸索一下内容就可以有效地理解和使用它。 然后，如果我必须使其起作用，就会开始出现问题-

• Is it possible or not? Has anybody done it before?有可能吗？ 有人做过吗？

• I was not sure which optical application to consider to use Machine Learning with.

  我不确定要与机器学习一起使用的光学应用程序 。

• Dataset- There are no open-source datasets available for any optical application. Again no big online optical community.

  Dataset-没有可用于任何光学应用的开源数据集。 再次没有大型在线眼镜社区。

• Coding help- Is there any initial code available to start with? Or do I have to write everything on my own?

  编码帮助-是否可以使用任何初始代码开始？ 还是我必须自己编写所有内容？

• The Biggest question- Is it really useful to use Machine Learning with optical, or I just wanted to use the buzz word technology in my work?

  最大的问题 -将机器学习与光学结合使用真的有用吗，还是我只是想在工作中使用流行词技术？

这个想法在风中消失了将近6个月…… (The idea lost in the wind for almost 6 months…)

With the above questions in mind, I gave up on the idea of using Machine Learning in Photonics. I guess this happens to everybody when they look for PhD topics. Every day, they come up with new ideas and then move on. Same happened to me and I got busy in typical optical/photonics application problems.

考虑到以上问题，我放弃了在光子学中使用机器学习的想法。 我想每个人在寻找博士学位主题时都会发生这种情况。 每天，他们提出新的想法，然后继续前进。 同样的事情发生在我身上，我忙于典型的光学/光子学应用问题。

It was a regular workday and maybe I was getting bored and started checking Google News. I exactly don’t remember. I saw a research paper getting published, which used Machine Leaning in some Photonic Power Splitters problem. And I remember saying this to myself- Oh Shit! You thought about it a few months back and it is possible… and it is not only about my excitement of using buzz word technology with optical applications.

这是一个正常的工作日，也许我感到无聊并开始查看Google新闻。 我完全不记得了 我看到了一篇研究论文发表，该论文在一些光子功率分配器问题中使用了机器倾斜。 我记得自己对我说过- 哎呀！ 您几个月前就考虑过，并且有可能……这不仅是因为我对在光学应用中使用流行词技术感到兴奋。

I can also use it and WILL USE IT…. All my doubts/questions were answered by itself…

我也可以使用它，并会使用它…。 我所有的疑问/问题都得到了回答……

最后的疑问... (Last Unanswered Doubt…)

Datasets?

数据集？

I still didn’t have any dataset available online. Then I thought if I find dataset online then I had to stick to that optical problem. But what if I can generate my own dataset depending on the problem I am interested in. This leads to a 1-month-long process of data collection on my own. It was around 10000 data-points, still far less than a general Machine Learning problem.

我仍然没有在线可用的任何数据集。 然后我想如果我能在线找到数据集，那我就不得不坚持那个光学问题。 但是，如果我可以根据自己感兴趣的问题生成自己的数据集，那该怎么办。这将导致我自己进行为期1个月的数据收集过程。 它大约有10000个数据点，仍然远远少于一般的机器学习问题。

What I want to say is if you are interested in a problem to use Machine Learning with, you can collect your own data. But again it depends on the type of problem you are interested in. In my case, it worked as I could collect it using our in-house developed code and fabrication facilities available in the research lab.

我想说的是，如果您对使用机器学习的问题感兴趣， 可以收集自己的数据 。 但这又取决于您感兴趣的问题的类型。就我而言，它可以正常工作，因为我可以使用研究实验室提供的内部开发代码和制造设施来收集问题。

一百万美元的问题… (A Million Dollar Question…)

I have been asked 3 times, during my final viva and 2 job interviews- Why to use Machine Learning with an Optical Application problem if we can use equations to predict or optimize a particular optical device?

在最后一次面试和两次面试中，我被问过3次了。如果我们可以使用方程式来预测或优化特定的光学设备，为什么要将机器学习应用于光学应用问题？

Machine Learning shines when there are a lot of input parameters to be optimized. First, if in our optical problem there are for example more than 10 input device dimensions to be optimized then we can easily employ machine learning. Comparing and optimizing 10 or more parameters one by one yourself is a tough job. Secondly, what if the data which we collected experimentally has some unknown noise factor which is never taken care of in equations. In this case, machine learning can be used to predict new values which will take into account the unknown noise of the experimental kit. So I think these 2 factors/cases make the application on machine learning in an optical problem very useful.

当有很多要优化的输入参数时，机器学习就会大放异彩。 首先，如果在我们的光学问题中，例如有10个以上的输入设备尺寸需要优化，那么我们可以轻松地采用机器学习。 自己一个人比较和优化10个或更多参数是一项艰巨的工作。 其次，如果我们通过实验收集的数据具有一些 未知的噪声因子，而该噪声因子在方程中从未涉及到呢？ 在这种情况下，机器学习可用于预测新值，该值将考虑实验套件的未知噪声。 因此，我认为这两个因素/情况使得在光学问题上的机器学习应用非常有用。

Later on, I published 2 journal papers out of this work and 1 was even selected for Editor’s Pick. Understanding Machine Learning which started as a curiosity for me ended up being the topic of my PhD. So you never know…

后来，我发表了2篇关于此工作的期刊论文，其中1篇甚至被选为《 编辑推荐》。 对机器学习的好奇开始于我的好奇心，最终成为我博士的主题。 所以你永远不知道...

As Steve Jobs said- You can only connect the dots looking backwards.

正如史蒂夫·乔布斯(Steve Jobs)所说-您只能将点向后看。

How to use Machine Learning in Optical, check the below article-

如何在Optical中使用机器学习，请查看以下文章-