911������

The Aalto Science Institute (AScI) International Summer Research Programme welcomed 68 talented students from around the world in 2024. Among them was Pengxin Wang, a master’s student from the University of Hong Kong, who spent three exciting months exploring Finland. Pengxin worked on a research project called Distilling Implicit Covariate Shift Induced by Node-Based Bayesian Neural Networks (BNNs) for Augmented Training. His goal was to make AI models not only smarter but also more reliable and efficient.

‘Deep learning based AI is often very accurate, but sometimes it can’t explain its decisions, or it struggles when the input data it sees changes,’ Pengxin explains. 'BNNs are great at solving this but are slow and require a lot of computer power.'

Pengxin designed a method to transfer knowledge into Deterministic Neural Networks (DNNs) that work faster while still being trustworthy.

The research group , where Pengxin conducted his project, is led by , who is deeply supportive of the AScI programme and is dedicated to mentoring young researchers. The group plays a key role in guiding interns to advance their careers and achieve significant breakthroughs in their fields.

With the invaluable guidance of Professor Kaski and PhD student Trung Trinh, Pengxin developed and tested his ideas. ‘I had the freedom to explore and experiment while getting great advice and support,’ he shares, reflecting on the collaborative and nurturing environment that defined his internship experience.

Starting the project wasn’t easy. ‘Before the project, I wasn’t very familiar with the field of Bayesian Deep Learning, and my coding skills needed improvement,’ Pengxin says. But with his advisor’s help and a lot of determination, he learned how to manage experiments, document his work, and present his findings clearly.

‘This was my first real research project in AI, and I learned so much,’ he says. ‘It taught me how to stay organized and handle challenges when things didn’t work as planned.’

This internship had a huge impact on Pengxin’s future. It inspired him to pursue a PhD in machine learning and continue working on AI that is not only smart but also reliable and trustworthy. He plans to stay connected with his advisor and collaborate on future projects.

Pengxin encourages students to apply for the program, even if they feel nervous about their abilities. ‘Choose a project that excites you and talk to the supervisors early. Everything else will fall into place because the program is so well-organized.’