Report from Singapore: "There is a huge interest in what we’re doing"

ICLR is considered one of the most important AI conferences in the world. In 2025, more than 11,000 papers were submitted, and 3,500 accepted. Two of them came from AI Sweden: Decoupled Subgraph Federated Learning and SWEb: A Large Web Dataset For The Scandinavian Languages.

A clear theme at this year's conference was AI Security – a topic highlighted in two of the opening keynotes. Johan Östman highlights presentations and articles on topics such as alignment, hallucinations, and building robust AI models.

"These are areas where we have already published many articles, and it's clear that we've made real progress. The most rewarding part of the conference was the networking. Several spontaneous conversations led to people from some of the best known labs asking to schedule longer meetings with us to learn more about projects like LeakPro," he says, and continues:

"I've been to many conferences before, but this was a new experience. In the past, I've been the one asking questions to top researchers to learn more about their work. This time, there was also a huge interest in our work."

The trip to Singapore also included several meetings with the Swedish Authority for Privacy Protection (Integritetsskyddsmyndigheten) and RISE, hosted by the Infocomm Media Development Authority (IMDA) – a government agency tasked with advancing Singapore’s digital economy and society – as well as the National University of Singapore.

"IMDA has been working with regulatory sandboxes for longer than we have in Sweden, and we gained a lot of insights from that. We also identified opportunities for collaboration in several of our ongoing projects," says Johan Östman.

Per Nydén, Acting Head of the Innovation Hub Unit at the Swedish Authority for Privacy Protection, summarizes his takeaways from the discussions in Singapore: