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

Deep neural networks (DNNs) have become the driving force behind recentartificial intelligence (AI) research. An important problem with implementing aneural network is the design of its architecture. Typically, such anarchitecture is obtained manually by exploring its hyperparameter space andkept fixed during training. This approach is time-consuming and inefficient.Another issue is that modern neural networks often contain millions ofparameters, whereas many applications and devices require small inferencemodels. However, efforts to migrate DNNs to such devices typically entail asignificant loss of classification accuracy. To address these challenges, wepropose a two-step neural network synthesis methodology, called DR+SCANN, thatcombines two complementary approaches to design compact and accurate DNNs. Atthe core of our framework is the SCANN methodology that uses three basicarchitecture-changing operations, namely connection growth, neuron growth, andconnection pruning, to synthesize feed-forward architectures with arbitrarystructure. SCANN encapsulates three synthesis methodologies that apply arepeated grow-and-prune paradigm to three architectural starting points.DR+SCANN combines the SCANN methodology with dataset dimensionality reductionto alleviate the curse of dimensionality. We demonstrate the efficacy of SCANNand DR+SCANN on various image and non-image datasets. We evaluate SCANN onMNIST and ImageNet benchmarks. In addition, we also evaluate the efficacy ofusing dimensionality reduction alongside SCANN (DR+SCANN) on nine small tomedium-size datasets. We also show that our synthesis methodology yields neuralnetworks that are much better at navigating the accuracy vs. energy efficiencyspace. This would enable neural network-based inference even onInternet-of-Things sensors.

SCANN: Synthesis of Compact and Accurate Neural Networks