Neurogenesis and multiple plasticity mechanisms enhance associative memory retrieval in a spiking network model of the hippocampus

Yansong Chua, Cheston Tan

Abstract

The hippocampal CA3 region is vital for long-term associative memory, yet most spiking network models remain small-scale proof-of-concepts. This study presents a biologically plausible, large-scale spiking neural network (SNN) of the hippocampus to investigate how neurogenesis and synaptic plasticity interact during memory encoding and retrieval. We demonstrate that while advanced synaptic plasticity is necessary for stable learning, structural plasticity—driven by neurogenesis and apoptosis—is essential to prevent network interference and manage memory overload.

Introduction: The Stability-Plasticity Dilemma

The hippocampal CA3 region is characterized by dense recurrent connectivity, allowing it to store memory traces known as engrams. However, computational models of this region face the "stability-plasticity dilemma": how to incorporate new information without overwriting existing memories or causing runaway network activity. Furthermore, the functional role of adult neurogenesis in the dentate gyrus (DG) and its impact on downstream CA3 memory processes has remained largely unexplored in spiking models. This research addresses these gaps by developing a large-scale SNN to study the interplay between synaptic and structural plasticity.

Methods: A Biologically Realistic Architecture

The researchers constructed a three-layer SNN comprising the Entorhinal Cortex (EC), Dentate Gyrus (DG), and CA3.

Schematic representation of the EC-DG-CA3 spiking neural network model.

Model Scale and Neuron Dynamics

The model utilizes over 4,000 Adaptive Integrate-and-Fire (AIF) neurons, representing a significant increase in scale over previous proof-of-concept models. This scale allows for a more realistic simulation of population dynamics and inhibitory feedback.

Plasticity Mechanisms

Three Spike-Timing-Dependent Plasticity (STDP) rules were evaluated:

  1. Pair STDP: Based on the timing between two spikes.
  2. Triplet STDP: Incorporates interactions between three spikes.
  3. Unified Plasticity: A sophisticated rule combining Hebbian learning with heterosynaptic and homeostatic mechanisms.

Simulating Neurogenesis

Structural plasticity was modeled in the DG by introducing new, highly excitable neurons while simultaneously removing older ones (apoptosis) to maintain a stable population. This process mimics the continuous turnover of neurons observed in the adult hippocampus.

Results: The Necessity of Structural Plasticity

The study revealed that synaptic plasticity alone is insufficient for managing high memory loads.

Plasticity Performance Comparison

The "Unified Plasticity" model significantly outperformed simpler STDP rules. As shown in Figure 5, the Unified (Full) STDP rule maintains high Signal-to-Noise Ratios (SNR) across a much broader range of synaptic weights compared to Pair or Triplet STDP.

Comparison of memory retrieval performance (SNR) across different plasticity models and network parameters.

Overcoming the Recruitment Trap

Without neurogenesis, the network falls into a "recruitment trap": each new memory stimulus activates an increasing number of CA3 neurons. This leads to excessive feedback inhibition that eventually suppresses the network's ability to retrieve any stored patterns.

Neurogenesis acts as a "refresh" mechanism. By introducing new neurons to capture recent stimuli and recycling old ones, the network maintains a stable recruitment rate. This allows for "graceful forgetting," where the system prioritizes recent memories while older, less relevant engrams fade, preventing total system failure during overload.

Discussion and Future Directions

The results suggest that structural plasticity is a requirement for functional memory systems when synaptic capacity is reached. While the current model provides a robust framework, future work could extend this to include the CA1 region for a complete hippocampal output pathway and investigate how neurogenesis aids in pattern separation for overlapping memory stimuli.

Summary updated

Abstract

Hippocampal CA3 is crucial for the formation of long-term associative memory. It has a heavily recurrent connectivity, and memories are thought to be stored as memory engrams in the CA3. However, despite its importance for memory storage and retrieval, spiking network models of the CA3 to date are relatively small-scale, and exist as only proof-of-concept models. Specifically, how neurogenesis in the dentate gyrus affects memory encoding and retrieval in the CA3 is not studied in such spiking models. Our work is the first to develop a biologically plausible spiking neural network model of hippocampal memory encoding and retrieval, with at least an order-of-magnitude more neurons than previous models. It is also the first to investigate the effect of neurogenesis on CA3 memory encoding and retrieval. Using such a model, we first show that a recently developed plasticity model is crucial for good encoding and retrieval. Next, we show how neural properties related to neurogenesis and neuronal death enhance storage and retrieval of associative memories in the CA3. In particular, we show that without neurogenesis, increasing number of CA3 neurons are recruited by each new memory stimulus, resulting in a corresponding increase in inhibition and poor memory retrieval as more memories are encoded. Neurogenesis, on the other hand, maintains the number of CA3 neurons recruited per stimulus, and enables the retrieval of recent memories, while forgetting the older ones. Our model suggests that structural plasticity (provided by neurogenesis and apoptosis) is required in the hippocampus for memory encoding and retrieval when the network is overloaded; synaptic plasticity alone does not suffice. The above results are obtained from an exhaustive study in the different plasticity models and network parameters.
BibTeX
@inproceedings{chua2017neurogenesis,
  author = {Yansong Chua and Cheston Tan},
  title = {{Neurogenesis and multiple plasticity mechanisms enhance associative memory retrieval in a spiking network model of the hippocampus}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2017},
  url = {https://openreview.net/forum?id=S7kGMbEe8QXV},
}
Figure