English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Convolutional Neural Networks (CNN) for image recognition tasks are seeingrapid advances in the available architectures and how networks are trainedbased on large computational infrastructure and standard datasets with millionsof images. In contrast, performance and time constraints for example, of smalldevices and free cloud GPUs necessitate efficient network training (i.e.,highest accuracy in the shortest inference time possible), often on smalldatasets. Here, we hypothesize that initially decreasing image size duringtraining makes the training process more efficient, because pre-shaping weightswith small images and later utilizing these weights with larger images reducesinitial network parameters and total inference time. We test this EfficientNetwork TRaining (ENTR) Hypothesis by training pre-trained Residual Network(ResNet) models (ResNet18, 34, & 50) on three small datasets (steelmicrostructures, bee images, and geographic aerial images) with a free cloudGPU. Based on three training regimes of i) not, ii) gradually or iii) in onestep increasing image size over the training process, we show that initiallyreducing image size increases training efficiency consistently across datasetsand networks. We interpret these results mechanistically in the framework ofregularization theory. Support for the ENTR hypothesis is an importantcontribution, because network efficiency improvements for image recognitiontasks are needed for practical applications. In the future, it will be excitingto see how the ENTR hypothesis holds for large standard datasets like ImageNetor CIFAR, to better understand the underlying mechanisms, and how these resultscompare to other fields such as structural learning.

Testing the Efficient Network TRaining (ENTR) Hypothesis: initially  reducing training image size makes Convolutional Neural Network training for  image recognition tasks more efficient