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

The process of optimizing the latency of DNN operators with ML models andhardware-in-the-loop, called auto-tuning, has established itself as a pervasivemethod for the deployment of neural networks. From a search space ofloop-optimizations, the candidate providing the best performance has to beselected. Performance of individual configurations is evaluated throughhardware measurements. The combinatorial explosion of possible configurations,together with the cost of hardware evaluation makes exhaustive explorations ofthe search space infeasible in practice. Machine Learning methods, like randomforests or reinforcement learning are used to aid in the selection ofcandidates for hardware evaluation. For general purpose hardware like x86 andGPGPU architectures impressive performance gains can be achieved, compared tohand-optimized libraries like cuDNN. The method is also useful in the space ofhardware accelerators with less wide-spread adoption, where a high-performancelibrary is not always available. However, hardware accelerators are often lessflexible with respect to their programming which leads to operatorconfigurations not executable on the hardware target. This work evaluates howthese invalid configurations affect the auto-tuning process and its underlyingperformance prediction model for the VTA hardware. From these results, avalidity-driven initialization method for AutoTVM is developed, only requiring41.6% of the necessary hardware measurements to find the best solution, whileimproving search robustness.

HW-Aware Initialization of DNN Auto-Tuning to Improve Exploration Time and Robustness