The brain's ability to process information through temporal sequences of spiking activity is crucial for complex cognitive tasks. These sequences arise from both sensory inputs and inherent network dynamics, spanning multiple, dynamically changing timescales. For instance, the syllables in the word ‘com-pu-ter’ would produce spike pattern sequences with different timings when spoken slowly or quickly, yet the semantic meaning remains invariant. What matters is the rank order of the sequence elements. Recent biological findings[1] and computational models suggest that dendritic plateau potentials offer a solution that bridges the gap between short, fixed time scale membrane potentials for coincidence detection and the long, timing-invariant processes needed for reliable sequence detection[2]. The proposed mechanism for dendritic sequence processing with plateau potentials is straightforward: Synchronous spikes activate localized plateau potentials in distal dendritic regions. This represents the detection of the first element in a sequence and depolarizes the neighboring dendrite segment towards the soma, enabling it to detect the subsequent sequence element by generating the next plateau potential. This cascade of activations progresses towards the soma, indicating the detection of an entire sequence. Despite its simplicity, this model has significant implications for neural computation: Firstly, it extends the expressivity of single neurons to robustly detect n-grams of spike patterns. As dendritic trees in the model branch, this expressivity further increases, enabling full branching expressions to be evaluated. Secondly, stochasticity of synapses has an outsized effect on cascades of all-or-none events. We will demonstrate how to harness stochasticity and produce a graded population level response that encodes the combined evidence for all sequence elements. Lastly, learning spatiotemporal patterns with significant timing variance is particularly challenging. We will present an initial approach to addressing this learning problem by identifying preceding plateau triggers, starting from the soma and tracing back in time. As evidence increasingly highlights the importance of dendrites and dendritic plateaus in complex behaviour[3], the model presented here offers a new perspective on neural computation that incorporates these insights and raises intriguing new questions we will highlight in the talk.