Antoine it’s great to meet you, can you please introduce yourself?
Sure, I am originally a Mathematician. I have a PhD in abstract Mathematics but 2 years ago I decided to move into the private sector and I did the ASI boot camp in data science here in London. After a few months, I was lucky enough to secure the position of Data Scientist at Voltaware, which was my first permanent data science job in the private sector.
At the beginning of my studies, I actually studied a bit of electrical engineering. I was always interested in Mathematics, but more interested in Physics. That is why I started in more engineering-focused studies. In France when you study engineering, you have a lot of mathematical training in your first 2 years, and I realised this was the part I enjoyed the most, so then I switched to pure mathematics.
So, it is ironic to come full circle to this field that mixes electrical engineering and machine learning!
You mentioned you focused your PhD on “abstract Mathematics”, can you give us some insight into what that was about?
Yeah sure, I was studying properties of boundaries of hyperbolic buildings. A hyperbolic building is an object you obtain by arranging simple pieces, like maybe triangles, according to some rather simple mathematical rules. When you arrange infinitely many pieces, even following simple rules, the resulting object can be very complicated and thus fascinating. In particular, the boundary of these objects is fractal and I was working on describing the geometric properties of these boundaries, it was a really good time!
It was a very fascinating topic, what I really liked is that geometry is doing mathematics with pictures in mind. As you go further into geometry, you are studying objects that are more and more complex, and some objects have infinite complexity. But you still need to find ways to visualise the aspect of the problem you are trying to solve. You have this unique object, and you have this question, then you need to find, in your head, a good point of view to solve your problem.
How has this mindset has prepared you for your career in Data Science?
In machine learning or data science, the visualisation of the data, can be the key to your problem so it is very similar!
If I look at Volatware’s data, we have 30 dimensions. Every point we have, lives in 30 different dimensions and obviously I cannot plot all 30 dimensions. I need find the dimensions that are the most important or I need to understand which transformation should be applied to the data to find a good visualisation of their features. This step is called feature selection in machine learning and is the key to make everything easier in the next few steps.
Feature selection is supported by good visualisation tools and is guided by the good mental pictures you can produce of your data.
When was the moment you decided to commit to the data science path, and what were your motivations for pursuing this?
I remember it very well actually! The first time I heard about data science was from my co-worker from my PhD studies. We shared the same office for 4 years. At one point I decided to focus on Mathematics, and do my post-doc, whereas he decided to move to the private sector. So, I saw his transition, and because I knew he was very happy with it, I got interested in data science. I asked him what I would need to learn to find a job and I started to realize I could be happy doing that because I would meet the intellectual challenges that I like.
At this time, I was actually a post-doc Israeli, so it has been quite a journey to get here. Then I decided to look for a job in data science and I did the boot camp in London because my wife was already working here, and I found personally that in data science, it was a great place to settle. At some point, we could have ended up in Paris, but I had the feeling London had more opportunities and my wife was very happy here; we were excited to discover life in London!
We see lots of amazing people who have conducted their research in Israel at some point, what is it about Israel that makes it stand out for this time of academia?
They have excellent Professors, and a very high calibre of students. In mathematics, the quality of Professors is due to immigration from Russia and the US, where you have stand-out Mathematicians.
About the students there, I think that one of the key factors is that in Israel you do 3 years compulsory military service. The benefit is that people start studying later, and are more mature, and more determined when they come to study difficult subjects. Whereas in France and the UK, people finish school and are sometimes unsure of what they want to do and have less determination.
What was the biggest challenge for you during your training to become a Data Scientist?
What I was afraid of was learning to code. I had learned to code at the beginning of my studies, but after 5-6 years in mathematics, I had not really touched a computer. Also, back then in engineering, we were mainly learning Java, and I realised in Data science Python is the thing. But I was surprised to see how simple Python is to learn. For people with no experience it is great. Little by little I overcame my problems, and I found it very stimulating! I enjoyed coding because of the mathematical structure you find in code. Good code is like a mathematical proof, it needs good definitions and good intermediate steps to be read and used by others.
The challenge when I was looking for a job, was to convince people I knew how to code. They saw me as an abstract person and were unsure about my ability to solve real-life problems. Having some coded projects is a big advantage in this situation.
If you are a good PhD student or post-doc, there is no reason why you cannot learn to code, and thus secure a role in data science.
If you had to pass on one piece of advice for someone looking to retrain to become a data scientist what would it be?
The main thing for me, is to apply what you learn in data science to a real-life problem. I would say to this person: “Find the topic or the challenge that excites you. There are tons of data sets and competitions available. Try to do it by yourself. You will understand some complexities of data science projects that are not necessarily explained in the textbooks. “
In terms of training, if you are not sure about your mathematical background, it is better to go back over the fundamentals of mathematics (linear algebra, calculus, probabilities and statistics). If you are confident about your mathematics, then just go over the main algorithms, understand how they work, how to implement them with the open-source libraries and try to solve real-life problems. Then you will be more passionate about the problems.
That leads us nicely into your current role! What was it about the opportunity at Voltaware that attracted you?
The key thing that I liked was that Voltaware has a scientific challenge to solve. We have data coming from our meters, and we need machine learning algorithms to deliver insights to our clients.
I prefer this, to let’s say data science applied to social science, where people are working in order to predict if you are going to buy the red pair of shoes or the blue ones. For me, the problems we are working on at Voltaware are much more exciting because they are more scientific. We can build scientific procedures to test and evaluate our models, so this was the main aspect that appealed to me.
Then, at this time, the company was really small, the data science team was only me and 2 Machine Learning Engineers, and so this was exciting, although stressful! It was my first job, and I was expected to deliver new models in just a few weeks, but I guess this was part of the excitement. To have a deep impact on the project, and to be able to see the output. If I had started in a big company, I would have worked with brilliant people, had lots of support, but I would have not seen the impact I did here.
Can you give us an introduction into Voltaware?
We are a smart metering company, we have sensors that analyse in almost real-time, the energy consumption in your house. And we are sending the data into the cloud, and in the cloud, we are running algorithms to deliver analytics to clients about their energy consumption.
We have 3 business cases. The first is energy optimisation. By giving our clients the breakdown of their cost by appliance and giving them intelligent recommendations about their usage, we can help them to reduce their energy bills. This is the project that has focused most of the data science team’s attention so far.
The second one is non-intrusive elderly care monitoring, because with this simple sensor, we can tell a lot of things about what is happening in your house. So, if you have an elderly relative living alone, maybe we can learn the pattern of habits of this person, and send a text message or alert, if their behavior changes drastically.
The third use case is appliance maintenance, the idea is to use this data to detect, in advance, the malfunction or break down in particular appliances. So, as a restaurant owner, if your fridge goes down you will lose ‘x’ amount of money. If we can tell you in advance when your fridge is behaving strangely, it can save you a lot.
What do you think the future of utilities within AI/Machine Learning looks like?
I think we are at the edge of a very exciting period because utility companies are usually big companies with huge inertia. Which is normal considering the time and investments required to produce electricity. Now with digital technologies, there are many opportunities for start-ups like us to lead the digital transition because utilities are very traditional and do not have a culture of quick innovation.
Innovations are urgently needed because in the next few years renewable energy and electric vehicles are going to impact massively the entire industry. Renewable energy production is very hard to predict. On the other hand, electric vehicles are going to double some utility bills. An increased demand associated with unstable production is going to lead to lower reliability on the grid. Digital technologies like Voltaware can have a key role in this situation to come.
As I said, the big players are quite static in this space, so there is a real opportunity for a start-up to breakthrough and have a huge impact. There is no Netflix of the digital utility revolution. Yet.
It is a great time to be joining Voltaware, you have just received significant funding from BP. Congratulations! What impact will that have on Voltaware in the near future?
A huge impact. The £1.5 million invested by BP Ventures will help us focus on our AI and data science capabilities. For the team, it is great news because it means we’re going to keep growing and that we will be able to work on new projects using Voltaware’s data.
In addition, credibility we get from our association with BP puts us in the spotlight. We have seen a huge rise in the number of inquiries since the announcement of the investment was made.
You have recently been promoted to Head of Data Science! How has the transition been to a slightly more hands-off role, and what advice would you give to someone making that transition?
The transition was quite a gradual one, as people were joining the team, which helped. We started at 3 and we are now 8 in the data science team, so it is natural that the structure of the team has evolved.
In this role, you have to guide the workflow of the team to meet the business requirements.
A key challenge for me, and the company, was finding the place where I could still be hands-on and checking code but also having enough time for the actual business side. So, I try and do half and half. 50% of my time is dedicated to coding and working on the projects that I initiated when the team was much smaller. The other 50% is dedicated to business, management, or product-related work.
For me personally it is important that I continue to code, because I like to solve problems, and I still have so much I can learn. I am working with very good Python and machine learning developers, and I just enjoy working and learning with them, discussing the technicalities with them.
I think maybe the best advice I can give is to keep your eyes and ears open to everything in the company. You have to listen to business, you have to listen to sales, and product people, and also listen to the guys coding with you because maybe the business is going to ask impossible things from the team. It is critical to have this overview and understanding of the company as a whole. You need to be disciplined in that 50/50 split, so you don’t lose sight of this.
Usually, business people, or sales people, don’t have a precise understanding of what machine learning can or cannot do, so it is in your hands, it is your task, as a data science lead, to drive the business. You might say “I can see this opportunity, with this data we can do ‘x’ very quickly, very simply and it is going to have ‘y’ impact on our product.”
That business understanding allows you to work with the sales and product team, instead of against. You are like a translator. I also spend a lot of time organising ways that the business and development team can work together, so we can increase awareness across the business, not just with me. For instance, I organised a Machine Learning seminar recently only for people outside of the team.
Inside our data science team, this is also my role to make sure that everyone understands our goals as a team. For instance, an infrastructure guy needs to understand, even partially, what the data scientists are doing and vice versa! This is getting harder as the team is growing but these exchanges are fruitful over the long term for the company.
In one word, please describe what you think makes a great data science leader?