Layerwise Learning-Rate Scaling in Vision Networks: Effective-Rate Identification, Power-Mean Aggregation, and Budget Dependence
Organizations: The authors are with the Computer Vision Lab, CAIDAS & IFI, University of Würzburg, 97074 Würzburg, Germany
Abstract
Layerwise learning-rate scaling assigns different learning rates to different parts of a network. Although established in fine-tuning pretrained models, its effects in vision networks trained from scratch are commonly assessed by training each candidate recipe. We show that, in this setting, the rate-related component of a layerwise strategy is well summarized by an effective learning rate computable directly from its multiplier profile. Across matched-control comparisons, changes in test accuracy closely track changes in training loss (r = -0.996), which supports an optimization-progress explanation of the short-budget rate effect. Increasing the untuned base learning rate tenfold shifts the rate-matched contrast upward in all six tested architectures (p = 0.016) and increases the number of architectures in which a published discriminative schedule, excluded from model fitting, outperforms the uniform control from two to five of six (p = 0.031). Within the evaluated strategy space, a parameter-weighted power mean with exponent 0.35 (95% confidence interval [0.20, 0.50]) accounts for 42.6% of within-architecture variance, compared with 36.4% for the arithmetic mean. Ranking candidates by this scalar without training improves the selected candidate's mean percentile rank from the 26th to the 10th and reduces its mean accuracy shortfall relative to the best candidate from 3.70 to 1.12 points. The rate-related effect decays as the training budget increases; relative to tuned baselines, the overall strategy effect falls within the prespecified practical-equivalence margin in two of three architectures. A separate grouping-related contrast can persist at long budgets, although its direction and magnitude depend on the architecture.
Figures & tables
| Rule | Evidence |
|---|---|
| Rank candidates by , , not by the arithmetic mean. | Zero-cost selection at the th percentile rather than the th; accuracy regret of rather than points; effect-sign AUC of . |
| Measure the rate sensitivity; do not predict it from the architecture. | Every descriptor set scores a negative leave-one-architecture-out . |
| Keep every group above the base rate. | points beyond the effective-rate cost once the mean is curved ( ): a threshold, not a gradient. |
| Decide the base rate first: a candidate’s sign depends on it. | Tenfold rise in lifts the pure rate cut in and ULMFiT from to . |
| Evaluate grouping at a matched mean rate and validate its effect at long budgets for each architecture. | cells at 10 epochs; points at 100 epochs on ConvNeXt but on GoogLeNet. |
| Assess the training budget before conducting a mean-rate search. | Decay as , to ; within the equivalence margin for tuned baselines. |
Layer-Wise Learning Rate Adaptation
Layer-wise learning rate adaptation assigns different or changing step sizes to model layers to improve training speed, stability, and generalization.
See all 17 papers