BCMfunktion

BCMfunktion is a term used in computational neuroscience and neural modeling to refer to the nonlinear function that governs synaptic modification in the Bienenstock–Cooper–Munro (BCM) learning framework. The function describes how a synapse changes its weight in response to presynaptic activity and the postsynaptic response, with the direction and magnitude of change depending on the current level of postsynaptic activity relative to a dynamic threshold.

The BCM theory, proposed by Bienenstock, Cooper, and Munro in 1982, posits that synaptic changes are activity-dependent and that the learning rule includes a sliding threshold that adapts to overall activity. The standard BCM function, often denoted φ(y, θ_M), is commonly written as φ(y, θ_M) = y (y − θ_M), where y is the postsynaptic activity and θ_M is a modifiable threshold.

In a typical BCM formulation, the synaptic weight update is proportional to the presynaptic activity x times the BCM function: Δw ∝ x φ(y, θ_M). The threshold θ_M itself evolves over time to track the long-term average of postsynaptic activity, helping to stabilize learning and prevent runaway potentiation or depression. Variants of the BCM function exist, including higher-order forms or different dynamics for θ_M, but the core idea remains: synaptic potentiation occurs when activity is above the threshold, and depression occurs when it is below.

BCMfunktion is used in simulations and theoretical analyses of unsupervised feature learning, sensory cortex modeling, and studies of metaplasticity. It provides a mechanism for adaptive, competition-driven changes in synaptic strength without external supervision, contributing to the emergence of selective, stable neural representations.