Burst Encoding hits 95.6% accuracy in biological neural classification

Researchers from multiple institutions, including Martin Schottlender and Veronika Volkova, evaluated how different encoding strategies affect closed-loop classification in cultured biological neural networks (BNNs). The team delivered visual inputs as spatiotemporal stimulation patterns via multi-electrode arrays (MEAs) and tested four encoding schemes: rate-based (spike frequency), phase-based (timing relative to oscillations), burst-based (clusters of spikes), and time-to-first-spike. In a binary classification task, burst-based temporal encoding dominated with 95.6% accuracy, while rate- and phase-based methods performed substantially worse.

The study, submitted to IEEE BioCAS 2026, also revealed that spatial distribution of stimulation is critical—suboptimal electrode selection could degrade accuracy significantly. The findings suggest that effective biological neural interfacing requires joint optimization of both temporal and spatial encoding strategies. This work underscores temporal encoding as a key design dimension for bio-digital computing, with potential applications in brain-computer interfaces and neuroprosthetics.