For the rest, ICCV is one of the biggest scientific conferences on Computer Vision, in 2019 taking place in Seoul, South Korea.
Why is this information relevant, though?
ICCV 2019 received 4303 papers, of which 1040 have been admitted, and 187 have been selected for public presentation in the conference auditorium filled with 6000 visitors.
That one time in 2019, among the hand-picked 187 papers, was the paper on “Task Routing”, written by Gjorgji Strezoski, a PhD candidate at the prestigious University of Amsterdam.
Besides finishing up his PhD., Gjorgji also teaches Fundamentals of Data Science, works as a Computer Vision researcher in the Rijksmuseum, and consults on AI-powered Art in Paradox StudiosNL.
Admissions are open for the Fall Data Science Brainster Bootcamp. Join us and future-proof your career by building a top-notch and job-ready portfolio.
We asked Gjorgji to give us an autograph and tell his career story 👇
Brainster: Hi Gjorgji, right off the bat — Python, or R😎?
Gjorgji: Objectively speaking, the scope of opportunities for data analysis is equally wide in both cases. However, my personal preference is Python.
The reason for that has nothing to do with any syntax or operational specifics. Python is simply an industry standard for functional systems.
An example: The easy integration with existing web platforms is a default. But the plentiful scientific literature offers Python code for testing bleeding-edge technology, a base for working with popular Deep Learning platforms like PyTorch, Keras, Lasagne, and the super-platforms below them like Caffe, Torch, and Tensorflow.
R, on the other hand, is more straightforward when it comes to statistical analysis, and offers a more transparent approach for data analysis based on maths and stats. It is good to know both, of course, but functionally, Python sounds more appealing, and we haven’t even touched the GPU support.
Brainster: In your paper presented in Seoul last year, you talked about Multitask Learning. Can you tell us more about that?
Gjorgji: Multitask Learning has to do with a kind of parameter partitioning in deep models so that they can support more than 1 task at a time.
A new kind of Multitask Learning — we called Many Task Learning — happens when a model can support more than 20 tasks at the same time. This is the main topic of the ICCV 2019 paper.
Brainster: As a Software Engineer, what made you switch to Data Science? Did you expect it to become the no.1 in-demand profession in the tech industry?
Gjorgji: I have a Bachelor of Science in Software Engineering. During my time at FINKI University, I had subjects that touched on data analysis and their structuring. Those subjects were not popular among the students at the time. When it comes to Data Science, there wasn’t a moment where I decided to switch. I started working on projects and slowly, gravitated towards them. I had no idea how important and necessary my skill would become.
At the time when I was working on MAESTRA (European project for analysis of unstructured big data), I noticed a big boom in the world of Machine Learning and Artificial Intelligence (~2014). Interesting architectures for neural networks appeared, along with even more interesting optimization principles. I had an opportunity to explore this field for the first time, and I consider it the final push forward towards my Data Science career.
Brainster: You’re teaching Fundamentals of Data Science at the prestigious UVA. What’s your opinion on learning Data Science through informal education?
Gjorgji: This part of the Data Science realm gives me mixed feelings. Partially, due to the big surge of courses and Bootcamps flying in from the furthest parts of the internet. Let’s be honest — with a personal blog anyone can be a teacher of anything.
In my opinion, for a quality informal education to be offered, a hand-picked selection of instructors is imperative, along with a transparent proof of qualifications, and an accordingly moderated curriculum. Additionally, in my experience having real-time and hands-on learning is the best way to learn because it enables discussion, clarification of questions, and gaining a thorough understanding of the subject matter — as opposed to informed intuition.
I salute Brainster’s initiative to enter the Data Science scene in Europe. I had the opportunity to gain insight into the program and I can confirm that it encompasses the crucial aspects that are needed for a successful career in Data Science.
Brainster: How does an average day of yours look like?
Gjorgji: I work on models that execute multiple tasks at the same time and formalize mutual primitives between the tasks they execute. This is a process led by the factor distribution of the data I work on. That’s why my main partner on this project is the Rijksmuseum in Amsterdam. The algorithms and models I develop are used for the analysis of artworks for the museum’s private use.
I have 4 workday prototypes:
#1: Typical working day at the institute. Starting with the new literature, reading through what’s new from the day before (1–2 hours). In my field, things are changing rapidly so every day I don’t catch up, which means having double material for the following day.
#2: Day at the museum. Much more relaxed than #1. Surrounded by art, new ideas, and great conversations. It starts with having a cup of coffee with art historians, photographers, and artists while catching up. This is an important part of my work because I draw inspiration from it.
#3: A day with partners. A stressful scenario where everyone who’d invested resources in my research, is trying to figure out how I’ve managed to progress. This day is filled with presentations, fiery discussions, and Eureka moments. If it goes well, I keep the resources and move on with the research… 🙂
#4: Teaching. This day consists of going to the auditorium, teach for 4 hours and do consultations for 3 more hours. My lectures are interactive and students can ask questions at any given moment. In the past 3 years, we built 6 successful projects from ideas born while discussing with students.
Join the next Fall Batch and start learning this month through the free Prep Programme. Master Data Science by working on real-life projects.
Brainster: In your opinion, how would the ideal path to a successful Data Science career look like?
Gjorgji: To begin with, a person must have an intuition about data. Formal certification is not mandatory for this, but it would be very useful. The knowledge gained at a technical University helps in understanding the concept of working with data. At the end of the day, Data Science is statistics and hands-on statistical analysis. Expanded with probability, calculus, the theory of groups, geometry, and cognitive science — it becomes so much more.
In my opinion, the ideal road to a career in Data Science means starting with a solid mathematical background. Continuing with a detailed understanding of the basic concepts about the basic algorithms being used, interesting projects that spark curiosity, constantly upgrading the skills and a deep understanding of the concept of data.
Brainster: Nothing is ever perfect, and Data Science as well. In your opinion, what’s the main problem with it?
Gjorgji: I think that the biggest problem currently is the sole definition of the term Data Scientist. It seems that at times a distinction between Data Scientist, Data Analyst, Data Engineer, Researcher, Research Engineer doesn’t exist. The job descriptions of the above mentioned are different, with some of the day-to-day activities orthogonal.
The reason for this is the recency of the emergence of DS as a distinguished field and the high, high demand for a skilled workforce.
Firstly, Pursuing a career in Data Science is not a warranty for a profitable career. A good Data Analyst earns much more and has a higher quality of professional life than an average Data Scientist.
Secondly, the other roles are important as well. If a Data Engineer doesn’t make a proper infrastructure for the Data Scientist to be able, to begin with, the analysis, Data Science wouldn’t exist. The bottom line is, a Data Analyst is absolutely necessary. To gain a deep understanding of the business logic and domain, as well as the needs of the clients. In that case, solid knowledge in Excel and pivot tables is superior to marginal knowledge of statistical analysis.
A clear distinction between all of the above-stated roles is necessary to make the most of the role in the workplace, and from there start building skills.
Brainster: What would you recommend to everyone who’s considering a career in Data Science?
Gjorgji: It’s never too late to invest in quality education. Time spent in gaining a deeper understanding of one’s profession is an investment with high ROI. If you’re right before the beginning of your DS career — and you’re curious — go for it! The least you can get is to understand the world you’re living in. You’ll be surprised when you find out how much data an average human creates, and how much control over you that holds.