Moonshine: Distilling with Cheap Convolutions
This addresses memory constraints for engineers deploying neural networks, but it is incremental as it builds on existing distillation methods.
The paper tackled the problem of deploying neural networks in memory-limited settings by proposing structural model distillation, which allows memory reduction without redesigning the teacher architecture. They showed substantial memory savings with minimal accuracy loss, achieving better performance than training the student directly on data, as demonstrated with residual networks on four benchmark datasets.
Many engineers wish to deploy modern neural networks in memory-limited settings; but the development of flexible methods for reducing memory use is in its infancy, and there is little knowledge of the resulting cost-benefit. We propose structural model distillation for memory reduction using a strategy that produces a student architecture that is a simple transformation of the teacher architecture: no redesign is needed, and the same hyperparameters can be used. Using attention transfer, we provide Pareto curves/tables for distillation of residual networks with four benchmark datasets, indicating the memory versus accuracy payoff. We show that substantial memory savings are possible with very little loss of accuracy, and confirm that distillation provides student network performance that is better than training that student architecture directly on data.