monad.ai
    Preparing search index...

    Variable LEARNING_RATEConst

    LEARNING_RATE: 0.01

    Gradient step size (α) applied per forward result.

    Small by design: Δweight = α × reward × contribution at α = 0.01 produces sub-1% shifts per step, keeping the weight trajectory smooth and preventing oscillation after a burst of unusual outcomes.