A Beginner’s Guide on Careers in Data Science in Japan

Getting ready for our upcoming Data Science bootcamp, we held a panel discussion with Data Science experts and Le Wagon teachers who shared their insights about the different career paths, skills needed to succeed and job prospects for Data Science in Japan.

Sebastien Beal is a machine learning expert by heart and entrepreneur by passion, running a Tokyo-based AI startup Locarise. His interest in building AI systems started from a childhood obsession with robots and their intelligence capabilities.  

Yann Le Guilly, Machine Learning Engineer at Arithmer Inc., excels in creating, developing and deploying AI-based products on the cloud, IoT and microcontrollers. Having a background in physics and maths, he enjoys mixing different aspects of science with software engineering. 

Trouni Tiet is a self-taught Data Scientist and tech entrepreneur who is leading our Data Science bootcamp at Le Wagon Tokyo. Already passionate about automation, witnessing the rise of AI and its emergence across all industries in the last decade has fueled his drive to join the field.

Yann: Data Scientist is a general umbrella term for all the data science-related jobs. I call myself a Machine Learning engineer because I want to stress out that I'm working in the stage of deployment models and services.

Sebastien: When I graduated from university, I was just an engineer that used different tools to solve a problem. Since the Data Science field has grown a lot, there is now enough material to build a full career in a specific topic. You start by learning the fundamentals and decide which sub-category you want to focus on. 

Trouni: Data Science is indeed very broad, and it's useful to experiment with a bit of everything to figure out what you like. For example, the Data Science bootcamp prepares our graduates for various industry roles across the production pipeline, ranging from data analysts to data engineers. However, data scientist roles with a strong research and theoretical component may require further learning, as the bootcamp strongly focuses on practical applications for the industry.

How does a Data Scientist’s daily work look like?

Yann: I work with structural data, so I spend less time on data and more on models, checking and changing them every day. If you do classical machine learning for unstructured data, you will need to devote more time to studying and processing the data itself. 

Sebastien: I work with data visualization a lot since we have retail customers who aren’t data scientists. I spend a lot of time trying to visualize data in a clear and simple way, and display results in a simple way for my clients to understand and make informed decisions. For example, most recently we had to analyze social distancing in shops and create heat maps. 

Yann: It's totally possible to work remotely. For deep learning, you often process data on the cloud or other computers which doesn’t change the flow much. In terms of security, you can deploy a VPN and all the typical security precautions that you can have on premises. 

Sebastien: All our engineers are working remotely, occasionally coming to the office once or twice a week. We actually just reduced our office size, so it was great timing.

Trouni: There are a lot of opportunities for English-only speaking technical teams. Most people need to have an understanding of written language because the latest research papers and documentation are written in English.   

Yann: For deep learning and most of the Data Science fields, you absolutely don't need to speak Japanese. However, you definitely need to speak Japanese if you work in the NLP field (natural language processing) to understand how your models are performing. If you work with unstructured data in Japanese, it can be hard but not a big issue overall. 

Which habits did you develop that have worked for you on your journey to becoming a Data Scientist?

Yann: Asking ‘Why?’. During data analysis, you need to research a lot about data, its origin and  why some parts are missing. When you start digging into these questions, you'll build more knowledge on top of that. 

Sebastien: Creating a research database. Labeling and organizing all the knowledge will help you to look back and search for some specific data. As you build up your skills over the years, you will realize that everything is connected, from probability models to deep learning. 

Yann: If you can spend hours digging into data in order to extract some specific knowledge, it could be a sign that you'll have fun working in Data Science. 

Trouni: Do you like working with Excel? Looking at spreadsheets may seem boring but if you enjoy crunching data out of it, you’re a good fit. 

Do I need to be an engineer or developer to get into data science? 

Trouni: Two years ago I would probably say ‘yes’ but my experience of teaching at Le Wagon has really changed my view on students’ capabilities. I have been repeatedly amazed by people coming from non-tech backgrounds who picked up things naturally even though they have never studied programming before. 

Trouni: I personally started my self-study by learning a lot of the theory behind machine learning. But looking back at it now, I would actually recommend beginners to start with a more practical approach. Don't be afraid to build things first, even though you don't understand the inner workings of your models. Machine learning requires to develop a lot of intuition about how to train models, and it only comes with real practice. Practical experience will also make it much easier to study and understand the underlying theory.

Is it possible for a sales and marketing professional to get into the business intelligence (BI) through learning data science? 

Sebastien: For BI, sales and marketing background is the best fit. I am always excited to see a person who can both understand and sell the technology at the same time. Go for it!

How to convince my company that I need to shift into Data Science? 

Yann: Finding some particular processes where you can include a data-driven decision can convince your company to sponsor your data science studies. The best way to get endorsed to get into data science is to pitch an idea of how this would help their business grow or be more profitable.

Yann: In my opinion, you should start with software engineering and learn to code first. While solving challenges, you will do a lot of research and gradually build this type of knowledge. 

Sebastien: The best way is to start with a problem you want to solve by coding. On top of that, having a basic understanding of algebra can accelerate your learning process. 

Which one should I learn, PyTorch or Tensorflow?

Yann: Don't learn specific frameworks, learn the concepts and it will be easier to apply in

different frameworks. If you know the structure of a convolutional network, you can easily implement it in Keras, MXNet, Tensorflow or PyTorch. 

Yann: Applied Data Science jobs don’t require any papers to be published. Now, if you want a pure researcher position, you will have to publish your papers in very famous journals. 

  

Sebastien: If you want to understand some specific approach, you might need to read fundamental papers related to the topic.

Sebastien: During interviews, companies often ask about how well you know the domain of their industry, so you will need to do a bit of research about it. The best match is to have the technical skill and the business knowledge of the industry in which you want to work.

Yann: Kaggle competitions are one of the best ways to stand out and gain the experience.

I also recommend working on projects you are interested in, adding it on your public github and emphasizing about it during hiring interviews. 

Trouni: If you don't have years of experience to show on your resume, then the best way to compete with those who do is to show concrete projects that you've built. Ideally, you can invest your time in one really good project that you feel passionate about. One good original project is better than five unoriginal ones done without much interest. Many recruiters look at what you do on the side because it shows how curious and dedicated you are about a topic.

Yann: There are only a few companies in the world that are successfully using reinforcement learning so it’s not an option for now.