The MADE Academy
What do you think people with PhDs do after they have written their dissertations and spent several years working on a project in production? Turns out, they recall the Latin proverb, “Scientia potentia est”, and continue learning!
Our employee Ekaterina, who has a PhD in Physics and Mathematics, challenged herself and spent a year intensively training with the MADE Big Data Academy, combining studies, full-time employment, and family matters. Here’s how she managed it.
Why did I choose the MADE Academy?
I heard about the MADE Big Data Academy from a colleague of mine, who is an experienced ML Engineer with a PhD. I thought to myself: “If he’s found something to learn over there, I should definitely check it out!”. Although I must confess, I also thought: “What am I, worse than him?” 🙂 That was the starting point of my year of studying Big Data Analysis concurrently with my workload, stress, and lack of sleep. Spoiler alert: I regret nothing! 🙂
How are exams conducted in MADE?
MADE has two tracks: Data Science and ML Engineering. The exams are identical for both tracks.
Enrolling in MADE is a serious undertaking. First you need to solve a series of Olympiad-level math problems within a limited time – these include differential equations, algebra, and logic problems. After that, you need to program a solution for algorithm tasks on CodeForces, and here you have limited time and resources. If your algorithm works but isn’t optimal – you have to redo it. The strongest contenders are offered to solve a Machine Learning problem. It consists of several stages and you get about 2 weeks to complete it.
How did I prepare for the exams?
“Pfft, what’s the big deal?” – I thought after watching a few YouTube videos on exam preparation for MADE (MADE itself provides a vast list of textbooks and courses you can use to prepare). And obviously… I didn’t pass. The main issue was the final interview, which I failed because I was speaking off the cuff.
The following year I decided to take a more serious approach – I rewatched the same videos and spent quite a bit of time on LeetCode. This time I said the right things to the interviewer – and here I am in MADE!
How did the studies go?
Back when I was taking the entrance exams, I was warned that it would be hard, so there were no real surprises. Studies took about 4 hours per day on weekdays and 8-10 hours on weekends, unless my body protested. Most of the time I spent watching lectures and doing homework. I aimed to watch to lectures live: firstly, it allows you to ask the lecturer as many questions as you need (and there were even bigger nerds than me); secondly, if you skip one, then your task debt will snowball in no time.
For the MADE class of 2022/2023, training was cut down to 1 year with minimal changes in the curriculum. Thus, the workload for each semester increased.
The first semester was organized in a way to let the students even out their ML and Python levels (Intensive Python course, Intensive Computer Science course, Machine Learning). It also included MLOps, Spark and Big Data, introduction to algorithms and data structures, advanced Python, and the defense of a class project. The semester was definitely no smooth sailing.
The class project is a story unto itself. We were broken up into teams and told that we had to get acquainted with the production and deployment processes of an ML/DS product and all accompanying factors, such as Business and Data Understanding, Model Development, Model Operations. Even though MADE students are generally responsible adults, every sprint was done in crunch mode. Each team had to develop a recommendation system for science papers, so there was also a competition among the projects (ours was in the top 3, by the way).
The second semester was even more intense, but, strangely, it got a bit easier. It may have been because of the pace we had set in the first semester. Sadly, not everyone was able to keep up, and by the second semester only 65 people out of 116 remained. The second semester included more specialized subjects, such as Deep Learning Basics, Advanced Machine Learning, Optimization Methods in Machine Learning, Deep Learning in Computer Vision, Natural Language Processing, А/В Testing, and, of course, our graduation projects, where teams worked on real business tasks under the mentorship of various Department Heads of the VK Group. Both the class project and graduation project were allowed to be presented at a project fair, where you could demonstrate your solution and meet potential employers.
A few words about the home tasks. Some of them were classic Python notebooks a la Coursera – complete the code and pass the assessment. Others were more complex, requiring you to prepare a dataset, realize neural network architecture from a science paper, and train it to a certain quality level. And there were also Kaggle competitions between MADE students. The main goal was to beat the baseline and, of course, your colleagues; the higher your rank on the leaderboard, the higher your grade! Math lovers also had a chance to solve some problems on methods of optimization with matrix-vector differentiation, assessment of function convexity, smoothness, and whatever else… It sounds weird, but almost everybody survived the second semester 🙂
Another unique aspect of studying in MADE is the community of teachers and students. The teachers are consummate professionals with a vast background of practicing what they teach. I especially enjoyed the lectures by Grisha Shevkoplias, Sergei Nikolenko (https://www.youtube.com/@snikolenko), and Radoslav Neichev (I hope I didn’t annoy them too much with my silly questions).
What did I have to sacrifice while studying?
From the very start everyone needed to answer to themselves what they have to forget about for a time in order to devote the lion’s share of their spare time to their studies. I firmly decided that it would never be to the detriment of my family. On weekdays I barely have time for them as it is, so on weekends it was my sacred duty to take my kid out for a walk, to the museum, the movies, the swimming pool or anywhere else that doesn’t have a laptop with unfinished homework or unwatched lectures.
What disappeared from my life: my few hobbies, sports (not that I spent much time on them in the first place), TV shows (although I did collect a nice backlog of stuff to watch), social media (no regrets on that one), sleep (this is a bit of an exaggeration: nobody will take away my nightly 7 hours of rest, but there were periods when I had to sacrifice some of it).
On the whole, a lot depends on the student’s level of training. Some have an easy time with algorithms, others know a million ways of choosing the best hyperparameters for training a network. Despite the arduous learning conditions, it was a ton of fun. I never regretted it for a second, and there wasn’t a single instance when I considered quitting.