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

Techniques for automatically designing deep neural network architectures suchas reinforcement learning based approaches have recently shown promisingresults. However, their success is based on vast computational resources (e.g.hundreds of GPUs), making them difficult to be widely used. A noticeablelimitation is that they still design and train each network from scratch duringthe exploration of the architecture space, which is highly inefficient. In thispaper, we propose a new framework toward efficient architecture search byexploring the architecture space based on the current network and reusing itsweights. We employ a reinforcement learning agent as the meta-controller, whoseaction is to grow the network depth or layer width with function-preservingtransformations. As such, the previously validated networks can be reused forfurther exploration, thus saves a large amount of computational cost. We applyour method to explore the architecture space of the plain convolutional neuralnetworks (no skip-connections, branching etc.) on image benchmark datasets(CIFAR-10, SVHN) with restricted computational resources (5 GPUs). Our methodcan design highly competitive networks that outperform existing networks usingthe same design scheme. On CIFAR-10, our model without skip-connectionsachieves 4.23\% test error rate, exceeding a vast majority of modernarchitectures and approaching DenseNet. Furthermore, by applying our method toexplore the DenseNet architecture space, we are able to achieve more accuratenetworks with fewer parameters.

Efficient Architecture Search by Network Transformation