Blog

Citizens of Silopolis: Anton Akusok

Anton Akusok is a senior machine learning expert focusing on advanced and high-performance computing. Anton has over five years of experience in machine learning projects, both as a project leader and as a researcher. 

Anton has a strong academic background: he studied at Aalto University in Finland and at the University of Iowa in the US. He received his PhD in Machine Learning and Big Data Analysis from the University of Iowa in 2016. 

During his studies Anton worked on a terabyte-scale distributed image analysis project in collaboration with one of the leading Finnish IT companies. His research focused on finding new applications for existing methods, covering cases like predicting incorrectly labeled samples in a dataset and data visualization with neural networks. Anton’s academic publications have been cited over 800 times.

Before joining Silo.AI, Anton published a GPU-accelerated randomized neural networks library called hpelm. He is also one of the founders of the applied machine learning program for continuous education at Arcada University of Applied Sciences, where he has been creating courses from scratch and taught several classes.

Understanding of high-performance programming helps with every project

In his research, Anton has focused on advanced and high-performance computing, something that he finds intrinsically fascinating:
“In my opinion, high-performance programming is one of the major challenges and the most fun one can have in computer science.”

High-performance computing aims at using exactly 100% of the computation power for useful calculations. This can happen either on a single computer, on a thousand computers at once, or on a heterogeneous environment including GPUs and specialized accelerators. Computer programs typically trade fast development time and convenience for performance; deep optimization is expensive to pull out, but it often improves the run speed significantly. Proper applications of high-performance computing are impressive, such as real-time speech recognition on your phone.

“The first thing someone new to the field would understand is that it's impossible to force your way with the code – a large poorly written program will overwhelm anyone. 'Work smarter, not harder' describes this field the best.” Anton says.

The challenge with high-performance computing is not only finding the slow part (for instance, reading files from a hard drive), but also studying and using the correct tools – often another programming language. To give an example, typically any demanding piece of code from Python is actually written in C++. 

Even so, the opportunities for applying high-performance computing are endless:
“In a visualization with neural networks I optimized a function from 100 to 5,000,000 evaluations per second on the same machine. The general understanding of writing high-performance code helps me with every project. My last project focused on large geospatial data analysis. In this case, I rewrote critical functions to be ten times faster with one third of the code. Work smarter, not harder!" Anton describes.

Tackling strenuous machine learning problems requires time and focus

Anton enjoys taking part in every stage of the project: from defining the opportunities to final delivery and deployment. He likes to see the results alongside the whole journey that the project has taken with the client.

“ In one case we brainstormed for the entire day with several AI experts from both Silo.AI and from the customer side. The results were very interesting, as the discovered opportunities were not directly related to the client's main business but also addressed a few of surrounding problems that caused delays and reduced the efficiency of work; a nice and unexpected discovery.” Anton comments.

When tackling strenuous problems, being able to concentrate is crucial. At Silo.AI we’ve built our organization and work environment to support our AI experts in their work:

“At Silo.AI, I enjoy having both challenging problems to solve and also having the time to focus on them and actually solve them, without many interruptions. Having the time to do your work properly sadly is becoming a luxury in the modern world, which is often too focused on cost-efficiency.” Anton comments.

“I like the freedom of work - no key cards to record your time, no daily reports, no following of instructions from above to the last letter. And the ability to propose and test new methods that can either fail or deliver outstanding results to a reasonable degree. This freedom must be the one of the keystone of AI excellence Silo.AI is able to offer." Anton says.

Another thing that Anton enjoys in his current work is the support he gets from his colleagues. After handling many university projects as a jack-of-all-trades, he appreciates the community and discussion with his team, his clients and the management. In any project, Anton is able to ask advice internally and often gets help from someone who has been through a similar technical problem.

With both the time and support needed, Anton feels like he is able to produce the best possible results with even with hard problems:
“I think the company does a phenomenal job creating a sustainable business with highly specialized skills of AI scientists. Usually a PhD scientist has two options: either to work as a researcher at a deep tech startup, or to take a software role that does not utilize one's full AI skillset. The interviews I've been to have proved that many companies find it difficult to hire scientists, as they simply don't know how to turn their deep scientific skills into value. Silo.AI is unique in a way that it brings these advanced skills into use for the real business and real world. The PhD researchers are just happy for the opportunity to earn a living by doing the thing we are best at and that we love the most.”

Carefully chosen area of expertise: advanced and high-performance computing 

Anton describes his career of becoming a highly specialized AI Scientist as a long path of focusing on the carefully chosen area, advanced and high-performance computing. Having turned down several offers and centering his attention on what he describes as “the beauty and simplicity of smart methods written in great code”, he has been able to become a better data scientist. This is telling of his ambition but also of his dedication to discover, develop and use these advanced technologies.

Anton outlined a couple of tips for aspiring data scientists to become better at their work:

  • One trick is to never go after "one best option". Instead, find all the available ways and reason with yourself which one you would take and why. You should even try to implement a few options and compare the results.
  • Never trust the code, especially machine learning code. There are a million things that can go wrong, and in programming the cost of fixing a mistake is always higher than the cost of writing good code from the beginning. 
  • Test your model and code constantly. Are you using a pre-trained deep learning network? Feed it an elephant picture, and check that it predicts an elephant. The red-green-blue pixels in your data may come in a different order than the network expects, and you would never know why the results are so bad otherwise. This is the biggest catch of machine learning: it cannot fail but it gladly gives out gibberish results. So never trust the code!

Games, nature and learning to be a dad

For Anton, computer games make the other side of the computer science medal:
"I'm not a professional player, but trying to stay ahead of the curve, if you know what I mean. Otherwise cycling and walking are nice ways to spend time with your thoughts and enjoy nature. At the same time learning how to be a dad – quite a challenge for which you cannot download a ready-made algorithm from Stackoverflow.”

Asking “why” as the most important Silo.AI value

“My favorite value of Silo.AI asking "Why?". Typically an AI project runs like a puzzle that we solve together with a client: fitting different AI methods, code algorithms, use cases, business impacts and feasibility limitations. Asking "Why?" to piece together a combination that brings value to our customers is the most thrilling part of my work; and the most rewarding as well.” Anton concludes.

Would you like to work with AI Scientists like Anton? We’re constantly looking for more curious AI experts. Get in touch with our recruiter jenni.kivikoski@silo-ai.hel5.wp-cloud.dev to discuss our current projects and check out our open positions.

Author
Pauliina Alanen
Former Head of Brand
Silo AI
Share on Social
Subscribe to our newsletter

Join the 5000+ subscribers who read the Silo AI monthly newsletter to be among the first to hear about the latest insights, articles, podcast episodes, webinars, and more.

Pauliina Alanen

Former Head of Brand

Silo AI

What to read next

Ready to level up your AI capabilities?

Succeeding in AI requires a commitment to long-term product development. Let’s start today.