AI 용어집
인공지능 완전 사전
BCM learning rule
Biophysical theory of synaptic plasticity introducing a plastic modification threshold that varies according to average post-synaptic activity. The BCM rule unifies LTP and LTD in a coherent mathematical framework.
Global synaptic plasticity
Coordinated synaptic modulation process at the neural network scale, often involving neuromodulators or reward signals. Global plasticity enables contextual adaptation and generalization in neuromorphic systems.
Metaplasticity
Phenomenon where the synaptic activity history modifies the future plasticity properties of the same synapse. Metaplasticity introduces longer-term memory of previous plastic states.
Structural plasticity
Modification of the physical connectivity of the neural network through creation, elimination, or restructuring of synapses. Unlike functional plasticity, it changes the topological architecture of neuromorphic circuits.
Functional plasticity
Variation in the efficacy of existing synaptic connections without modifying the network's topological structure. Functional plasticity primarily concerns changes in synaptic weights.
Oja's rule
Learning algorithm that modifies Hebb's rule to stabilize synaptic weights by introducing a normalization term. Oja's rule enables principal component extraction in neuromorphic networks.
Frequency-dependent plasticity
Form of plasticity where the direction and amplitude of synaptic modification depend on the stimulation frequency. It generally distinguishes low frequencies (inducing LTD) from high frequencies (inducing LTP).
Calcium-dependent plasticity
Mechanism where intracellular calcium concentration determines the direction of synaptic modification. Low concentrations favor LTD while high concentrations induce LTP.
Generalized Hebbian Rule
Extension of the Hebbian principle including synaptic depression and normalization terms to avoid explosive weight growth. This more realistic formulation is widely used in neuromorphic implementations.
Triplet-STDP Plasticity
Variant of STDP that considers interactions between three action potentials to better reproduce biological experimental data. It provides more accurate modeling of temporal learning phenomena.
Spike-Bundle Plasticity
Plasticity mechanism where groups or "bundles" of action potentials, rather than isolated spikes, determine synaptic modifications. This approach better captures the dynamic nature of neuronal transmission.