SOURCE-LINKED INTELLIGENCE

Theoretical Guarantees for One-Shot Magnitude Pruning and Compute-Adaptive Early Exit

arXiv · AI, language, vision and robotics · article · Sep 11, 2026 · UTC

We study compute reduction in neural networks through a unified partial versus full computation view, captured by one-shot magnitude pruning in the static regime and early exit in the adaptive regime. In an asymptotic single-neuron model, we prove a concentration theorem for one-shot magnitude pruning with explicit rates. We also introduce the conditional perceptron for early exit and show that its excess generalization error decays as a power of the compute gap, with an exponent that grows to infinity as the alignment between partial and full computations tends to one. We then extend the anal

Read original source ↗ Open in workspace

recordType

paper

region

Global