It is symbolic that Seoul was chosen as the venue—a city at the epicenter of the global hardware race (recall the recent South Korean $576 billion investment plan). ICML traditionally sets the technology stack for the next few years. This year, the focus of participants has shifted from the extensive scaling of LLM parameters to inference optimization, model interpretability (white boxes), Edge AI, and resolving topological bottlenecks. What is discussed on posters in Seoul today will become the standard for B2B integration and commercial APIs tomorrow.
Source: ICML
R&DICML 2026SeoulMachine LearningDeepTech