Systematic Study Ranks Learning-Rate Schedulers Across 30 Architectures

arXiv·medium signal

Classical AutoML often treats the learning-rate scheduler as a secondary hyperparameter, but this work systematically measures its impact on classification accuracy across a diverse pool of 30 convolutional and transformer architectures from the LEMUR neural-network dataset, using automated source-code instrumentation. The result is empirical guidance on scheduler choice rather than manual guesswork. Practical for practitioners tuning training runs across heterogeneous model families.