An automatic software algorithm to generate new synthetic neural training data for BCI/BMI systems using information maximizing diffusion models.
The brain–computer interface (BCI) market is rapidly expanding as demand grows for systems that translate neural activity into actionable information across healthcare, assistive technology, research, and human–machine interaction. Despite this growth, the market faces persistent challenges related to the high cost, time, and effort required to collect sufficient neural and behavioral data to build reliable, scalable solutions. Variability in data over time and the limited availability of diverse training examples increase development costs and slow deployment, creating strong demand for approaches that can reduce data collection burdens, improve system stability, and accelerate commercialization across the BCI ecosystem.
Researchers at Emory and Georgia Tech aim to increase the effective dataset size available for BCI systems by leveraging recent advances in diffusion modeling to produce novel and diverse examples of neural activity. This capability expands opportunities across both scientific and engineering efforts in systems neuroscience and BCI-based rehabilitation. The technology directly addresses key limitations in the field, including shifts in recording conditions over time and the scarcity of training data across diverse tasks, which traditionally require repeated data collection and system recalibration. By capturing the structure and variability of neural population signals, this approach has the potential to support the development of improved decoding and behavioral inference methods using fewer real measurements, reducing the burden of data collection and facilitating the development and deployment of scalable BCI solutions.
Proof of concept on synthetic data available.