Methods And Systems For Efficient Processing Of Recurrent Neural Networks
AssigneeApplied Brain Research Inc.
InventorsGurshaant Singh Malik, Aaron Russell VOELKER, Christopher David ELIASMITH
Life of this patent
Estimating the term…
Loading patent…
AssigneeApplied Brain Research Inc.
InventorsGurshaant Singh Malik, Aaron Russell VOELKER, Christopher David ELIASMITH
Estimating the term…
Recurrent neural networks are efficiently mapped to hardware computation blocks specifically designed for Legendre Memory Unit (LMU) cells, Projected LSTM cells, and Feed Forward cells. Iterative resource allocation algorithms are used to partition recurrent neural networks and time multiplex them onto a spatial distribution of computation blocks, guided by multivariable optimizations for power, performance, and accuracy. Embodiments of the invention provide systems for low power, high performance deployment of recurrent neural networks for battery sensitive applications such as automatic speech recognition (ASR), keyword spotting (KWS), biomedical signal processing, and other applications that involve processing time-series data.
A digital hardware system for processing time series data with a recurrent neural network, wherein the digital hardware system comprises a plurality of computation blocks, the digital hardware system computes parts of a given recurrent neural network by time multiplexing the parts of the recurrent neural network over a spatially distributed set of computation blocks of the digital hardware system;
A method for producing a digital hardware system for processing a recurrent neural network to balance requirements of energy, latency, throughput and accuracy, that accepts as input:
Dependencies are inferred from the claim text.