A day in the life of is a series that champions Tech voices in the APAC region, to inspire anyone looking to begin a career and/or pivot into Tech.
TL;DR | This is for you if you’re
- Looking to find out more about working in a computer vision Tech startup
- Wondering what it’s like pursuing a career in research first before starting your career in data/AI-related roles
- Learning about the behind-the-scenes of AI
Rebekah: Hey Jeroen! Thanks so much for saying yes to our interview today.
Jeroen: No worries. It’s always cool to hang out with interesting people and mix with them coming out of two years in lockdown.
Rebekah: Absolutely and I’m so excited to kick this off. So tell us a bit about yourself — I read that you studied in Amsterdam?
Jeroen: That’s right. Now you probably can hear from my accent that I’m Dutch so it’s not a big surprise that I ended up at university there. I studied business information systems — which is kind of a mixed degree of computer science and business with the organisational side of business. I then did a PhD, which was more focused on computer vision (CV), now it’s called AI but back then there wasn’t such a thing.
Rebekah: What part of computer vision interests you — what made you want to focus on CV in particular?
Jeroen: That’s a good question. So images in particular, they’re very unstructured information. And I have this urge to find out how I can structure that information. And I always say wherever I’ve worked is basically, I want to make sure that people don’t have to see all the images.
Rebekah: What do you mean by that?
Jeroen: You know how many millions of hours of YouTube are being uploaded as we speak? I still want you to see images but only those that really matter to you.
Rebekah: That’s fascinating. So what does this mean in terms of your day to day?
Working in a computer vision Tech startup
Jeroen: This means for my job, I have to have an understanding of what these images mean. So it’s the most important part — for humans to act on more information, connect, and test our understanding of the data. This is what I studied.
Rebekah: Yes, I understand that you were in the research, academia side of things for a while. What was that like?
Starting off in academia and research vs going commercial
Jeroen: So the nice thing about academia is that you get the opportunity to delve very deeply into matters and really think about them without any commercial pressure. So it’s great especially when you start your PhD.
Rebekah: What did you enjoy about it?
Jeroen: I love the invention part of it, doing things that nobody has done before — so I’m still doing it. But I also love innovation which simply means the practical use of inventions in my case.
Rebekah: What’s the real difference between doing research and going commercial?
Jeroen: If you work in the commercialisation of R&D — which is what I’ve been doing for the past few decades — then you focus much more on the real issues than when you’re in academia.
Rebekah: I see. So what would you recommend — starting off in research or going straight to work in computer vision?
Jeroen: Ah, well — you’re going to get a biassed view from me, you know right?
Rebekah: Haha yes!
You’re interested in computer vision — where do you start?
Jeroen: So firstly yes, if you want to jump into computer vision — start with your hobby projects. Go and jump in! Compared to my time, technology wasn’t so readily available. Now, everything is at your fingertips. My teenage son is just grabbing things from the Internet and running neural networks. If you can just do those things — fantastic, do that! You can even start companies from it. But don’t forget to get the fundamentals — the foundational education.
Rebekah: Absolutely, that’s what I love about the Tech industry. There are no walls, or barriers to entry. We don’t need a professional qualification per se. We can just learn to code online! And the community plus support are incredible.
Jeroen: Yes, and I do see, it’s very tempting. Now you have all these wonderful micro degrees and I think they’re great for people who have the foundations. But you shouldn’t do it instead of the foundations. And I think if you’re a student now it’s going to be super exciting — way more exciting than in my time! And back then it was already exciting, because we were kind of witnesses to the birth of quantum computing.
Rebekah: Wow, big words. Please explain!
Jeroen: For example AI applications — nobody was thinking about that 10 years ago. But you can only get into that now if you have some level of foundation. So don’t skip on that. And when life moves on to whatever is next like the metaverse, you can just say, “Okay, well, now I have to do a micro degree on top of my foundations and I’ll be good to go”, but you have to have a basic understanding of that first. Does that make sense?
Rebekah: Yes it does! So what would you say is a “good foundation” to have?
Jeroen: So I’m not going to be too excited about data science!
Rebekah: Haha ooh boy. Tell me more!
Computer science and statistics still a top choice
Jeroen: I think you will be much better off with, well, preferably a mixed degree of computer science and statistics. In the end, most of data science is statistics. Computer science obviously helps you apply that and the missing part, that is more about how to apply it in business. So I guess that will come forward in the more specialised data science courses. But again, the business side — that’s probably the easiest part to learn on the job. But statistics and computer science are not.
Rebekah: Any other course of study you might recommend?
Jeroen: Hmm. So at ProofTec we do some work with insurance and I speak with data scientists — a lot of whom have studied actuarial science. That’s actually quite a good foundation too. It’s a little bit specialised but it might actually be a much better study than pure data science.
Rebekah: Ah yes. I recently interviewed James, the Head of Data Science at Datium, and he started off in physics!
Jeroen: That’s nice. Software is very specialised now. When I was teaching at Sydney University, I had a lot of students from architecture. I was very surprised! And they were actually among my best students. They were not learning computer science, but learning architecture gave them part of the toolkit to do well.
What it’s like being CTO of a CV startup
Rebekah: That’s so cool! Alright, Jeroen, could you tell me a bit about your day to day as CTO of ProofTec?
Jeroen: I knew this question was coming from Steve. So in my mind, my day to day might not be too exciting, so I’ll tell you more about the week I’ve had.
Rebekah: Ah, love Steve! Sure thing, tell me about your week.
Jeroen: In the past few days, I’ve been talking to our investors about the company’s strategy. So what should we focus on next? How do we match our core strengths to market opportunities and realise our potential? It’s actually quite an intellectual exercise. So that was all very high level. But last week, what I did was completely the opposite. So I was deep down into the launch of one mobile app and an update on another mobile app. Also, a web app update, hardware installation designs for several sites and discussing some very specific R&D issues with the team. So there’s a lot of variety in being a CTO. And of course this is a startup, so I have to do everything of course.
Rebekah: Yes, we all wear many hats at startups!
Jeroen: Yes, and that makes the job a bit hard sometimes because when you’re deep down into things it’s much harder to switch than when you’re at the high level. But it also makes it very exciting because you get exposed to all parts of the business.
Rebekah: Absolutely. I love the pace of being in a startup and how every day is different. How did you go from your PhD to the role you have now?
Jeroen: There’s a lot of coincidence involved here. So what happened is, when I moved to Australia I was offered two jobs. One was at a commercial research institute and the other was at a university. And somehow something went wrong with the HR department — I ended up with two jobs! That’s when I decided to do the teaching at university and move my research to the commercial institute.
Rebekah: Oh what was the thought process behind that?
Being in research vs commercial
Jeroen: The main reason for that was funding. What happens in research is you really have to fight for your funding every year, while in a company, while that is also true in some sense, there’s more continuity and the equipment available to you is also much better. Of course there are downsides but I really enjoyed being more on the commercial side and being much more connected to what happens with the research. So long story short, I kind of half stumbled into the commercial side of things and I liked it — so I stayed on.
Rebekah: A very nice coincidence! So tell me a little bit about ProofTec and the problems you’re solving?
Rare events and damage detection in computer vision
Jeroen: So, at a very high technical level, we specialise in looking at rare events. So these are statistically rare — meaning you basically get lots of data points where nothing happens. And that’s very difficult.
Rebekah: What sort of rare events does ProofTec look at?
Jeroen: Now we’re working on damage detection. So chances of that happening is much higher than zero but still pretty small — maybe 2% of the cases have damage. And how do you find out? It’s the needle in or it’s a couple of needles in a haystack. So that’s one thing. The other thing that we specialise in is finding very small things, very small anomalies.
Rebekah: When you say small things or noise, how small is small?
Jeroen: Our first product has been finding damage on rental cars. Small scratches that require 25 times zoom. People can see them but in practice they don’t because it’s a rare data point.
Rebekah: Interesting. What’s the use case for this in the real world?
It’s the interaction paradigm. Because AI produces probabilities only, we have to help humans turn the probabilities into actionable intelligence.
Jeroen: So maybe if you buy a second-hand car, you as a person would go over it. But if you’re a car rental company and 100 cars come in per day, you can’t look at each car with the same level of detail. People might say, it’s not technical but I think it is. It’s the interaction paradigm. Because AI produces probabilities only, we have to help humans turn the probabilities into actionable intelligence. Which is very hard because people are our users, and they don’t think in terms of probabilities. But probabilities that come out of an AI cannot be truly rounded. So we spend a lot of time figuring out how to interact with users differently.
Rebekah: What do you mean by figuring out how to interact with users differently?
Jeroen: Depending on the use case, the exact same AI results can be presented in a different way. For example, one use case may need binary decisions, accepting a performance trade-off. While another use case is better served with triage, basically a ranking. On top of that, we have worked with an UX expert to, in his words, “humanise the AI”, translating numbers with ten decimal places to something that users can emotionally connect to. This is work in progress, and an understudied field of research.
Rebekah: That is so fascinating. I think humanising AI will be pivotal. Another thing that plays a huge role in machine learning is the labelling of data. I would love to know what role data labelling play in your business?
The role of data labeling in computer vision
It’s essential. So the kind of anomaly detection that we do requires quite strict supervised learning because there is a lot of subjectivity. Labelling is so important because it helps to organise the subjectivity and diversity in the data.
Jeroen: It’s essential. So the kind of anomaly detection that we do requires quite strict supervised learning because there is a lot of subjectivity. Labelling is so important because it helps to organise the subjectivity and diversity in the data.
Rebekah: What do you mean by diversity?
Jeroen: So, people always talk about the quantity of the data, but variety in the data is especially important, and you need to label all of that.
Rebekah: Why is that so?
Jeroen: We’re trying to automate humans, right? So we have to get all that information from humans into it. And humans know so much. They’re generally intelligent. And these AI’s have specialised intelligence, they are not generally intelligent. So, this is why a finger in front of the camera might look like a rare event because our AI has no concept of fingers, it only looks at cars. There are no fingers in cars but a human would immediately see it and know that it must be a finger.
Rebekah: Ah, I got it. Okay so just to wrap up the session before we run out of time — what advice would you give to someone wanting to start their career in AI/ML?
Career advice for those who’re just starting out in AI/ML
Jeroen: First of all, you should definitely do it! It’s such an interesting area and I think the first choice you have to make is whether you want to become a researcher or a practitioner. So basically work with all the available tools that are here now and the new ones that will be developed or do you actually want to bring AI to them? Do you want to build machine learning models or do you want to deploy and integrate these models with an end product?
And the other thing that’s important is you need to be able to converse about probabilities. You have to be able to talk about probabilities with other people that you work with — your team members, customers, other stakeholders. So just understanding it is not enough. If you can speak about probabilities, then you can actually build great products and have great results, not just to create data science. And for me AI, it doesn’t just mean artificial intelligence, but it also means actionable intelligence.
If you can speak about probabilities, then you can actually build great products and have great results, not just to create data science. And for me AI, it doesn’t just mean artificial intelligence, but it also means actionable intelligence.
Rebekah: Wow, that’s so good. But speaking about probabilities doesn’t sound easy.
Jeroen: I’m still learning. I was actually talking to somebody a few months ago who was much more experienced in it than I am — I’m still learning from that.
Rebekah: I love that. So important to have a growth mindset and to keep on learning. I think that’s all the time we have for today, Jeroen. Thank you for spending 30 minutes with me!
Jeroen: All right, fantastic.
Find out more about Jeroen on LinkedIn here.
Have any questions you might want to know from Tech voices across Asia? Let us know in the comments or write to rebekah@supa.so 🚀
