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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Bayesian Neural Networks (BNNs) that possess a property of uncertaintyestimation have been increasingly adopted in a wide range of safety-critical AIapplications which demand reliable and robust decision making, e.g.,self-driving, rescue robots, medical image diagnosis. The training procedure ofa probabilistic BNN model involves training an ensemble of sampled DNN models,which induces orders of magnitude larger volume of data movement than traininga single DNN model. In this paper, we reveal that the root cause for BNNtraining inefficiency originates from the massive off-chip data transfer byGaussian Random Variables (GRVs). To tackle this challenge, we propose a noveldesign that eliminates all the off-chip data transfer by GRVs through thereversed shifting of Linear Feedback Shift Registers (LFSRs) without incurringany training accuracy loss. To efficiently support our LFSR reversion strategyat the hardware level, we explore the design space of the current DNNaccelerators and identify the optimal computation mapping scheme to bestaccommodate our strategy. By leveraging this finding, we design and prototypethe first highly efficient BNN training accelerator, named Shift-BNN, that islow-cost and scalable. Extensive evaluation on five representative BNN modelsdemonstrates that Shift-BNN achieves an average of 4.9x (up to 10.8x) boost inenergy efficiency and 1.6x (up to 2.8x) speedup over the baseline DNN trainingaccelerator.

Shift-BNN: Highly-Efficient Probabilistic Bayesian Neural Network Training via Memory-Friendly Pattern Retrieving