Sebastian Szyller: I want to make machine learning more dependable and resilient to attacks

What do you research?

My research focuses on trustworthy and adversarial machine learning. It's a cross-disciplinary research field at the intersection of machine learning, security/privacy, and broadly defined transparency. I explore ways of making machine learning-based systems more dependable and more resilient to malicious cyber attacks

For instance, I’ve studied how to train privacy-preserving machine learning models, how to safeguard them and ensure robust predictions in untrusted environments.

Why are you interested in this topic?

Since my undergraduate days, I have had a keen interest in large-scale data systems.

Initially, I got into big data and shortly after that, machine learning. I quickly realised that in order to be broadly deployed and reach their full potential, machine learning models would require an immense amount of safeguards to increase their trustworthiness.

From there, I spent almost ten years investigating various aspects of the field.I began with a bachelor’s thesis at the intersection of machine learning and cryptography. This was followed by a joint research-focused MSc and PhD with a strong focus on the defenses against model stealing and ownership verification.

Eventually, while working as a research scientist at Intel Labs, I examined the security and privacy of large language models.

My industry experience improved my ability to clearly define and scope projects. It also, gave me a unique dual perspective: Co-authoring proposals as a researcher, while also evaluating academic proposals from the other side.

These experiences shaped my vision of effective academia-business collaborations.

What is the most important quality of a researcher?

To me, research projects resemble the startup scene quite a bit. They involve many ideas, validations and iterations.

There are two key factors to consider. From a principled perspective, researchers have to be curious to invent or discover new things–embracing an unapologetic vision, if you will. However, researchers also have to be quite industrious.

Science progresses rapidly, and you need to stay on top of it. Additionally, you need to wear many hats: supervision, funding, outreach, to name just a few.

What are your expectations for the future?

Establishing a successful lab is a no-brainer. But a loftier goal would be to establish trustworthy machine learning as a key competence in Finland.

We have great faculty that cover fundamental security, machine learning and adjacent areas. I want to build on this foundation by creating a hub for trustworthy adversarial machine learning research that attracts top talent and fosters collaboration across institutions (both academia and industry).