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 achieved tremendous success for cognitiveapplications. The core operation in a DNN is the dot product between quantizedinputs and weights. Prior works exploit the weight/input repetition that arisesdue to quantization to avoid redundant computations in Convolutional NeuralNetworks (CNNs). However, in this paper we show that their effectiveness isseverely limited when applied to Fully-Connected (FC) layers, which arecommonly used in state-of-the-art DNNs, as it is the case of modern RecurrentNeural Networks (RNNs) and Transformer models.  To improve energy-efficiency of FC computation we present CREW, a hardwareaccelerator that implements Computation Reuse and an Efficient Weight Storagemechanism to exploit the large number of repeated weights in FC layers. CREWfirst performs the multiplications of the unique weights by their respectiveinputs and stores the results in an on-chip buffer. The storage requirementsare modest due to the small number of unique weights and the relatively smallsize of the input compared to convolutional layers. Next, CREW computes eachoutput by fetching and adding its required products. To this end, each weightis replaced offline by an index in the buffer of unique products. Indices aretypically smaller than the quantized weights, since the number of uniqueweights for each input tends to be much lower than the range of quantizedweights, which reduces storage and memory bandwidth requirements.  Overall, CREW greatly reduces the number of multiplications and providessignificant savings in model memory footprint and memory bandwidth usage. Weevaluate CREW on a diverse set of modern DNNs. On average, CREW provides 2.61xspeedup and 2.42x energy savings over a TPU-like accelerator. Compared to UCNN,a state-of-art computation reuse technique, CREW achieves 2.10x speedup and2.08x energy savings on average.

CREW: Computation Reuse and Efficient Weight Storage for Hardware-accelerated MLPs and RNNs