Machine Learning with Sami – A Technical Insight
28.04.22
The things we do are Ancient Hebrew for many people. Yet, machine learning is very much part of our everyday life. Imagine; search engines, vacuums that move about independently around your house (dogs are usually afraid of these, cats sit on top of them), and cars driving around without human hands on the steering wheel. Your three matches on Tinder last weekend, and your lazy-ass friend with only two. The list goes on and on… Machine learning is something very, very exciting, and it is something that can be used for solving many complex products and to develop cool new products.
Our aim in Vaisto is to stay on top of these new technologies. Unfortunately, the writer of this interview doesn’t know much about machine learnings principles, but luckily, we have one of Vaisto’s machine learning gurus at hand.
Sami Dahlman, what’s up?
– There’s spring in the air, so all good!
You will have to disappoint our readers and me because as you can’t share everything because of project confidentiality, but give us a glimpse of what you do.
– I’m lucky to be working with something I’m very passionate about. Digitalisation is one thing that changes the business landscape for all our customers. That in turn opens up exciting opportunities to leverage machine learning and analytics as the tools of trade.
– What comes to machine learning, data is the starting point. At start, the data is often pretty raw and unprocessed – sometimes even incomprehensible, needing a proper “clean-up”. The data in this form is not yet ready for machine learning training, but a hefty amount of preprocessing is needed. A close and continuous dialogue with customer domain specialists is crucial, too. Through iterative data preprocessing, discussions, and machine learning training rounds, the machine learning model and the application gradually become real.
What do you think the trends look like?
– I hope 80’s would come back with band t-shirts and long hair…
Sorry… but how about trying to focus on the machine learning trends for a bit?
– That’s a big topic. If you ask me, I’m pretty sure that the hunt for Artificial General Intelligence (AGI) continues. The actual real-life AGI is still years away, but investments in this area bring concrete results for the #AI community to benefit from.
– There are clear signs that the machine learning based solutions are becoming more reliable and are better supporting specialists in various industries in making critical and time sensitive decisions. One could say that machine learning applications have grown out of research labs.
– For many end-users, the machine learning predictions are still somewhat a black box – something miraculous that comes out, but difficult to comprehend how it was actually done… End-users expect a more transparent view of machine learning application operation. This is clearly an area that needs improvement.
– Building the datasets for machine learning applications still takes a lot of effort. The direction is that there are more tools for supporting dataset preparation; from AI-assisted coding, synthetic dataset generators to semi-supervised data annotation tools for building your custom models.
Can you talk us through a typical project? Given that some bright mind wants to join us?
– Machine learning projects come in many shapes and sizes, depending on what kind of customer problems need to be solved. This of course makes our work all the more enjoyable. In general, our workflow is the following:
Customer data status vs. customer target application review
Data exploration and preprocessing
Dataset construction and initial model runs based on agreed methods
MLOps pipeline setup for model optimization rounds
Best model deployment to production, and
Production time model updates based on new data collection
Technology-wise, what do you see as the next big things?
– The level of autonomy is growing practically in all areas, whether it’s mobility, robotics, or home electronics. AI-driven specialist support systems are becoming more flexible, cognitive and context-aware better supporting specialist tasks.
So this is briefly what we do. If you want to take a deeper look in Machine Learning and become one the gurus, join us.
Text by Kristian
