View source for Meta-plasticity

'''Meta-plasticity''' refers to the plasticity of plasticity: the rules governing how a neural system learns are themselves modifiable by experience. In standard synaptic plasticity, individual synaptic weights change according to fixed rules (e.g., Hebbian or STDP protocols). In meta-plasticity, the parameters of those rules — learning rates, thresholds, time constants, the balance between potentiation and depression — are themselves dynamical variables that change with the system's history.

The concept arises from the observation that biological synapses do not have fixed learning rules. A synapse that has been recently potentiated becomes harder to potentiate further and easier to depress — a phenomenon called synaptic tagging and capture. This is not a single fixed rule but a context-dependent modulation of the learning rule itself.

Meta-plasticity introduces a second level of structural-dynamical coupling: the structure (synaptic weights) changes through plasticity, and the plasticity rules themselves (the meta-structure) change through experience. The system is not merely learning. It is learning how to learn.

The engineering analogue is '''meta-learning''' in machine learning, where a model learns an update rule that generalizes across tasks. However, meta-learning typically operates on explicit parameter spaces, while biological meta-plasticity operates on the molecular machinery of the synapse — a much more constrained but also more robust substrate.

Meta-plasticity is thought to be a mechanism for stability-plasticity dilemma: a system that is too plastic forgets; a system that is too stable cannot adapt. Meta-plasticity may regulate the tradeoff dynamically, increasing plasticity when the environment changes and decreasing it when the environment is stable.