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

Long Short-Term Memory (LSTM) is widely used in speech recognition. In orderto achieve higher prediction accuracy, machine learning scientists have builtlarger and larger models. Such large model is both computation intensive andmemory intensive. Deploying such bulky model results in high power consumptionand leads to high total cost of ownership (TCO) of a data center. In order tospeedup the prediction and make it energy efficient, we first propose aload-balance-aware pruning method that can compress the LSTM model size by 20x(10x from pruning and 2x from quantization) with negligible loss of theprediction accuracy. The pruned model is friendly for parallel processing.Next, we propose scheduler that encodes and partitions the compressed model toeach PE for parallelism, and schedule the complicated LSTM data flow. Finally,we design the hardware architecture, named Efficient Speech Recognition Engine(ESE) that works directly on the compressed model. Implemented on XilinxXCKU060 FPGA running at 200MHz, ESE has a performance of 282 GOPS workingdirectly on the compressed LSTM network, corresponding to 2.52 TOPS on theuncompressed one, and processes a full LSTM for speech recognition with a powerdissipation of 41 Watts. Evaluated on the LSTM for speech recognitionbenchmark, ESE is 43x and 3x faster than Core i7 5930k CPU and Pascal Titan XGPU implementations. It achieves 40x and 11.5x higher energy efficiencycompared with the CPU and GPU respectively.

ESE: Efficient Speech Recognition Engine with Sparse LSTM on FPGA