anti-Hebbian learning

the inverse of Hebbian learning: correlated activity weakens the connection. neurons that fire together lose their shared weight.

$$\Delta w_{ij} = -\eta \cdot x_i \cdot x_j$$

anti-Hebbian plasticity serves as an inhibitory signal. where Hebbian learning concentrates representation (amplifying co-occurring patterns), anti-Hebbian learning decorrelates representation (suppressing redundancy). the result: sparse, efficient codes where each neuron carries independent information.

found in the cerebellum (parallel fiber to Purkinje cell synapses), the hippocampus (feedforward inhibition), and in independent component analysis (ICA) — a computational model that recovers statistically independent sources from mixed signals.

in cyber

market inhibition is the anti-Hebbian mechanism in the cybergraph. the inversely coupled bonding surface (ICBS) suppresses cyberlinks the collective disbelieves — edges with low market-implied probability are weighted toward zero. karma penalizes neurons whose links consistently lose market confidence.

$$A^{\text{eff}}_{pq} = \sum a(\ell)\cdot \kappa(\nu(\ell))\cdot f(m(\ell))$$

when $m(\ell) \to 0$ (market rejects the link), $f(m(\ell)) \to 0$ — the connection is suppressed. this is anti-Hebbian: correlated rejection weakens the edge.

the ternary triad

anti-Hebbian learning is the inhibitory (-1) member of the three irreducible learning types: Hebbian learning, anti-Hebbian learning, homeostatic learning. excitation, inhibition, modulation — the ternary architecture of intelligence. see two three paradox.

see learning, synaptic plasticity

Local Graph