Building a realistic, scalable memory model with independent engrams using a homeostatic mechanism

Abstract

AbstractMemory formation is usually associated with Hebbian learning, using synaptic plasticity to change the synaptic strengths but omitting structural changes. Recent work suggests that structural plasticity can also lead to silent memory engrams, reproducing a conditioned learning paradigm with neuron ensembles. However, this work is limited by its way of synapse formation, enabling the formation of only one memory engram. Overcoming this, our model allows the formation of many engrams simultaneously while retaining high neurophysiological accuracy, e.g., as found in cortical columns. We achieve this by substituting the random synapse formation with the Model of Structural Plasticity. As a homeostatic model, neurons regulate their own activity by growing and pruning of synaptic elements based on their current activity. Utilizing synapse formation with respect to the Euclidean distance between the neurons with a scalable algorithm allows us to easily simulate 4 million neurons with 343 memory engrams. These engrams do not interfere with one another by default, yet we can change the simulation parameters to form long-reaching associations. Our model paves the way for simulations addressing further inquiries, ranging from memory chains and hierarchies to complex memory systems comprising areas with different learning mechanisms.Author summaryMemory is usually explained by the strengthening or weakening of synapses that fire closely after each other. This limits the forming of memory to already connected neurons. Recent work showed that memory engrams can be formed with structural plasticity on a homeostatic base. In this model, the synapses of neurons have a fixed weight, and the model forms new synapses to have sufficient synaptic input to maintain a steady fire rate. However, this model connects neurons uniformly at random, limiting the parallel formation of memories. We extend this work by using a more biologically accurate model that connects neurons in a distant-dependent manner enabling us the parallel formation of memories. The scalability of our model enables us to easily simulate 4 million neurons with 343 memories in parallel without any unwanted inference. Our model increases biological accuracy by modeling brain-like structures such as cortical columns and paving the way for large-scale experiments with memory.