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

DL compiler's primary function is to translate DNN programs written inhigh-level DL frameworks such as PyTorch and TensorFlow into portableexecutables. These executables can then be flexibly executed by the deployedhost programs. However, existing DL compilers rely on a tracing mechanism,which involves feeding a runtime input to a neural network program and tracingthe program execution paths to generate the computational graph necessary forcompilation. Unfortunately, this mechanism falls short when dealing with moderndynamic neural networks (DyNNs) that possess varying computational graphsdepending on the inputs. Consequently, conventional DL compilers struggle toaccurately compile DyNNs into executable code. To address this limitation, wepropose \tool, a general approach that enables any existing DL compiler tosuccessfully compile DyNNs. \tool tackles the dynamic nature of DyNNs byintroducing a compilation mechanism that redistributes the control and dataflow of the original DNN programs during the compilation process. Specifically,\tool develops program analysis and program transformation techniques toconvert a dynamic neural network into multiple sub-neural networks. Eachsub-neural network is devoid of conditional statements and is compiledindependently. Furthermore, \tool synthesizes a host module that models thecontrol flow of the DyNNs and facilitates the invocation of the sub-neuralnetworks. Our evaluation demonstrates the effectiveness of \tool, achieving a100\% success rate in compiling all dynamic neural networks. Moreover, thecompiled executables generated by \tool exhibit significantly improvedperformance, running between $1.12\times$ and $20.21\times$ faster than theoriginal DyNNs executed on general-purpose DL frameworks.

DyCL: Dynamic Neural Network Compilation Via Program Rewriting and Graph Optimization