“Control devices with thought alone” ... Hanyang University independently develops the world’s top-performing “EEG AI”

[Financial News] South Korean researchers have independently developed the world’s top-performing electroencephalogram (EEG) artificial intelligence (AI) model, which can control computers or robots with thought alone and precisely diagnose brain disorders. With a foundation in place that could significantly lower the barriers to commercializing EEG technology, efforts to secure a leading position in the BCI (brain-computer interface) and digital healthcare industries are expected to accelerate.
Hanyang University announced on the 13th that a research team led by Professor Lim Chang-hwan of its Department of Biomedical Engineering had developed SPERA, an EEG AI foundation model that can be used across a range of EEG applications. The findings were accepted for presentation at NeurIPS 2026, one of the world’s most prestigious AI conferences, and will be formally presented in Sydney, Australia, this December.
A foundation model is an AI system pretrained on large-scale data to build a foundation of capabilities. Just as a student with a solid grounding in basic subjects can easily learn a range of other subjects, an AI that has already learned the brainwave patterns of many people can quickly and accurately perform a variety of tasks—including sleep analysis, emotion recognition and brain disorder diagnosis—even with limited data.
The research team trained SPERA on EEG data from approximately 29,000 people and nearly 80,000 hours of EEG recordings. After testing its performance against nine publicly available global benchmarks, the team found that SPERA ranked first in eight evaluation areas, surpassing existing world-leading models, and achieved an average accuracy 3.52 percentage points higher. In particular, its accuracy in “imagined speech”—identifying words thought of internally from EEG signals—was 8.69 percentage points higher than that of existing models.
From an industrial standpoint, a major competitive advantage is the potential to dramatically reduce the time and cost required for technology development. Previously, developing EEG AI required collecting and processing large datasets from scratch each time. But in a sleep-stage analysis task, SPERA used just 5% of the total data and achieved higher accuracy than existing AI trained on 100% of the data.
The achievement is expected to serve as a shared platform for a range of commercial neuroengineering technologies, including robotic control for patients with quadriplegia, software for the early diagnosis of brain disorders, and devices for stress and sleep management.
Hanyang University Professor Lim Chang-hwan said, “SPERA, which we developed independently, has significantly outperformed existing models on global performance benchmarks,” adding, “It will become a standard AI model that leads the way in EEG AI research worldwide and the development of industrial applications.”  

[email protected] Kim Man-gi Reporter