Remembering like a Brain: A Bio-Inspired Associative Memory Model integrating Hodgkin-Huxley Dynamics and Modular Network Architecture

\\Associative memory enables the retrieval of stored information from incomplete or noisy cues and supports cognitive functions such as pattern recognition, decision-making, and learning. Computational models, particularly Hopfield-type networks, have provided valuable insights into the basis of these processes. However, many existing models overlook important biological factors, including realistic neuronal dynamics, synaptic variability, and inhibitory mechanisms, which limits their neurobiological plausibility. \\Building upon the model by H. Hasegawa (2001), we propose a novel associative memory model that integrates biologically realistic Hodgkin-Huxley-type neurons within a modular Hopfield network architecture. Our model incorporates the three predominant neuron types found in the mammalian brain, facilitating the simulation of complex network interactions such as lateral inhibition and disinhibition. This biologically inspired framework supports both chemical and electrical synapses, dynamic connectivity, and a modular network structure, thereby enhancing the biological realism and computational efficiency of memory process modeling. \\Our network consists of a three layered architecture of $N$ extended spiking Hodgkin-Huxley neurons with time-delayed couplings that store $P$ patterns through their synaptic weights. \\Our simulations show high network performances with an accurate biological fidelity, achieving a storage capacity ($\alpha_c = P_{max}/N \sim 0.7$) for low neuron activity ($f=0.1$), which represents a significant improvement compared to previous associative memory models. \bigskip\noindent{\bf Keywords}: Associative memory, Hodgkin–Huxley model, Hopfield network, graph theory, computational models, heterogeneous firing patterns, synaptic coupling.