Active dendrites form the basis for biologically plausible neural networks possessing desirable features of the biological brain including flexibility, adaptability, and high energy efficiency. A formulation for active dendrites using the notational language of conventional machine learning is put forward as an alternative to a spiking neuron formulation. Based on this formulation an online clustering unit (OCU) is developed as a basic neural building block, and its capabilities are demonstrated via an application from experimental neuroscience: spike sorting.

Spike sorting takes inputs from electrical probes embedded in neural tissue, detects voltage spikes (action potentials) emitted by neurons, and attempts to sort the spikes according to the neuron that emitted them. Many spike sorting methods form clusters based on the shapes of action potential waveforms, under the assumption that all spikes emitted by a given neuron have similar shapes and will map to the same cluster. Clustering is challenging because there are natural variations both between different neurons and spike instances emitted by the same neuron leading to significant overlaps in the spike shapes.

Using synthetic spike shapes, the accuracy of the proposed OCU is compared with a much more compute- intensive, offline k-means approach. The OCU outperforms k-means and has the advantage of requiring only a single pass through the input stream, learning as it goes. The overall capabilities of the OCU are demonstrated for a number of scenarios including dynamic changes in the input stream, differing neuron spike rates, and varying cluster counts.