Salvato in:
Dettagli Bibliografici
Autori principali: Korol, Guilherme, Beck, Antonio Carlos Schneider, Castrillon, Jeronimo
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2505.17626
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915300826415104
author Korol, Guilherme
Beck, Antonio Carlos Schneider
Castrillon, Jeronimo
author_facet Korol, Guilherme
Beck, Antonio Carlos Schneider
Castrillon, Jeronimo
contents Dynamic DNN optimization techniques such as layer-skipping offer increased adaptability and efficiency gains but can lead to i) a larger memory footprint as in decision gates, ii) increased training complexity (e.g., with non-differentiable operations), and iii) less control over performance-quality trade-offs due to its inherent input-dependent execution. To approach these issues, we propose a simpler yet effective alternative for adaptive inference with a zero-overhead, single-model, and time-predictable inference. Central to our approach is the observation that models trained with Stochastic Depth -- a method for faster training of residual networks -- become more resilient to arbitrary layer-skipping at inference time. We propose a method to first select near Pareto-optimal skipping configurations from a stochastically-trained model to adapt the inference at runtime later. Compared to original ResNets, our method shows improvements of up to 2X in power efficiency at accuracy drops as low as 0.71%.
format Preprint
id arxiv_https___arxiv_org_abs_2505_17626
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Leveraging Stochastic Depth Training for Adaptive Inference
Korol, Guilherme
Beck, Antonio Carlos Schneider
Castrillon, Jeronimo
Machine Learning
Hardware Architecture
Dynamic DNN optimization techniques such as layer-skipping offer increased adaptability and efficiency gains but can lead to i) a larger memory footprint as in decision gates, ii) increased training complexity (e.g., with non-differentiable operations), and iii) less control over performance-quality trade-offs due to its inherent input-dependent execution. To approach these issues, we propose a simpler yet effective alternative for adaptive inference with a zero-overhead, single-model, and time-predictable inference. Central to our approach is the observation that models trained with Stochastic Depth -- a method for faster training of residual networks -- become more resilient to arbitrary layer-skipping at inference time. We propose a method to first select near Pareto-optimal skipping configurations from a stochastically-trained model to adapt the inference at runtime later. Compared to original ResNets, our method shows improvements of up to 2X in power efficiency at accuracy drops as low as 0.71%.
title Leveraging Stochastic Depth Training for Adaptive Inference
topic Machine Learning
Hardware Architecture
url https://arxiv.org/abs/2505.17626