A neural theory for counting memories

"Ive never smelled anything like this." "Ive seen you once before." "Ive heard this song many times." Keeping track of the number of times different stimuli have been experienced is a critical computation for behavior. This computation occurs ubiquitously across sensory modalities, and naturally without reward or punishment. However, the neural circuitry that mediates this computation remains unknown. Here, we propose a theoretical two-layer neural circuit that can store counts of stimulus occurrence frequencies. This circuit implements a data structure, called a count sketch, that is commonly used in computer science to maintain item frequencies in streaming data. Our first model implements the count sketch data structure using Hebbian synapses and outputs stimulus-specific frequencies. Our second model uses anti-Hebbian plasticity and only tracks frequencies within four count categories ("1-2-3-many"), which we suggest makes a better trade-off between the number of categories that need to be distinguished and the potential ethological value of those categories. Using real-world datasets, we show how both models can closely track the frequencies of different stimuli experienced, while being robust to noise, thus expanding the traditional novelty-familiarity memory axis from binary to continuous. Finally, we show that an implementation of the "1-2-3-many" count sketch -- including network architecture, synaptic plasticity rule, and output neuron that encodes count categories -- exists in a novelty detection circuit in the insect mushroom body, and we argue that similar circuit motifs also appear in mammals, suggesting that basic memory counting machinery may be broadly conserved.