Dendritic excitatory-inhibitory balance and branch-specific gating enable selective recall of associative memories

The ability to reconstruct complete memories from fragmented cues is a defining feature of associative memory, which requires neural circuits to store many overlapping and conflicting memory representations while selectively controlling memory accessibility. Attractor dynamics provide the dominant modelling framework for associative memory and have been used to interpret connectivity and behaviour. Yet despite this influence, how attractor-based memory is implemented in biophysical spiking circuits with compartmental dendrites and structured recurrent interactions has remained unresolved, with no unifying theoretical framework for interpreting emerging dendritic imaging and connectomic data. Here we show that local dendritic excitatory-inhibitory balance creates binary-like membrane-potential states that enable associative recall, and that branch-selective inhibition gates access to memory sets stored across dendritic trees, limiting interference between distinct sets. We further show that continuous dendritic dynamics enlarge basins of attraction by integrating partial cues over time, and that a winner-take-all readout circuit implements memory-set selection autonomously, generating interneuron activity that encodes set identity. Our framework predicts branch-specific voltage and calcium signatures that provide a basis for interpreting dendritic measurements in terms of engram organisation, bridging classical associative memory theory with dendritic circuit mechanisms.