Why we felt like doing this
A couple of weeks ago, ITP received a mail from Google's partner program, allowing us to do the Professional Machine Learning Certificate at no cost. If it weren't for this invitation, it would probably have taken a couple more months before feeling confident enough to endeavor in this ML adventure. After all, the requirements include three years of experience in Machine Learning on GCP and a thorough understanding of the AI components that it offers. Two conditions that I didn't meet.
Nevertheless, together with Kenny Helsens, we decided to go for it and challenge ourselves.
In addition, we were motivated to showcase our ML expertise and contribute to our OKRs on thought leadership and to be acknowledged as a premium partner in Machine Learning for our clients.
Preparing for the exam
Preparing for the exam wasn't straightforward. From the resources provided by Google, it was not that clear what the exam's focus would be. In hindsight, our focus was biassed too much towards the technical knowledge, which the exam only skimmed over.
I would recommend focusing on all the AI components, solutions, products, and APIs that GCP offers. The questions often query your knowledge of infrastructure and technologies in GCP to best solve the problem.
The exam page offers some sample questions. These questions represent difficulty, but not of scope, as there is just a vast amount of products that Google offers, each with its pros and cons depending on the use-case.
We discovered a lot of use-cases and resources which helped us, which we documented on a Miro board. Other people have written elaborate blog posts on what ground to cover while preparing for the exam. Read through all of them, and you’ll get a pretty good impression of the scope of the exam. The board also contains a list of YouTube videos from TensorFlow Developer Summit and Google I/O that provide a good starting point in new topics like AI Fairness, KubeFlow or TFX.
Feel free to use these resources as a guide! Good luck!
Performing the exam
Timing, timing, timing.
The biggest challenge during the exam was keeping track of time. The exam consists of 60 questions you have to answer in 120 minutes. Virtually all questions present a use-case/scenario, described in 1 paragraph. Most questions are single choice with four options. Because of the nature of each question, it takes quite a lot of time to assess all the possibilities. A handful of questions started from a code sample that you had to read and understand to come up with the correct answer!
Although the phrasing and the options are pretty revealing to what the correct answer should be, it is crucial to keep in mind to use the most straightforward solution that demands the least amount of effort and leverages the AI and Data solutions from GCP most.
To conclude, I believe that the professional ML certificate exam aims to assess your general understanding of AI and ML. Furthermore, it offers a good evaluation of ML intuition as it demands you to evaluate a wide range of scenarios.
On the other hand, the certificate fails to evaluate deep knowledge of ML and AI. Most of the resources listed above come from the Google documentation catalog and cover advanced ML technologies in GCP. However, the exam does not query that knowledge directly and focuses on how the products function on a higher level.
Nevertheless, I believe that the exam has the benefit of creating a baseline for ML knowledge at ITP. Furthermore, it offers an excellent perspective for new joiners in the AI team and showcases our capabilities to clients.
Feel free to reach out to the AI team if you have any questions about the ML exam!