Zum Hauptinhalt springen
UB Siegen

FLEXIBLE HARDWARE FOR HIGH THROUGHPUT VECTOR DEQUANTIZATION WITH DYNAMIC VECTOR LENGTH AND CODEBOOK SIZE

2018
Online Patent
Zugriff:
Zum Volltext

The performance of a neural network (NN) and/or deep neural network (DNN) can limited by the number of operations being performed as well as memory data management of a NN/DNN. Using vector quantization of neuron weight values, the processing of data by neurons can be optimize the number of operations as well as memory utilization to enhance the overall performance of a NN/DNN. Operatively, one or more contiguous segments of weight values can be converted into one or more vectors of arbitrary length and each of the one or more vectors can be assigned an index. The generated indexes can be stored in an exemplary vector quantization lookup table and retrieved by exemplary fast weight lookup hardware at run time on the flyas part of an exemplary data processing function of the NN as part of an inline de-quantization operation to obtain needed one or more neuron weight values.

Titel:
FLEXIBLE HARDWARE FOR HIGH THROUGHPUT VECTOR DEQUANTIZATION WITH DYNAMIC VECTOR LENGTH AND CODEBOOK SIZE
Link:
Veröffentlichung: 2018
Medientyp: Patent
Sonstiges:
  • Nachgewiesen in: USPTO Patent Applications
  • Sprachen: English
  • Document Number: 20180300603
  • Publication Date: October 18, 2018
  • Appl. No: 15/881519
  • Application Filed: January 26, 2018
  • Claim: 1. A system for enhanced data processing in a neural network environment (100), the system comprising: at least one processor (205); at least one memory component (220, 225); and at least one memory (210) in communication with the at least one processor, the at least one memory (210) having computer-readable instructions stored thereupon that, when executed by the at least one processor, cause the at least one processor to: receive one or more initialization parameters from a cooperating controller component (220) of the neural network environment (100), the initialization parameters comprising data representative of the dimensions of the data (340) to be processed by the neural network environment (100) and data representative of one or more vector quantization index values (615), the one or more index values representative of one or more vectors (625(a), 625(b), 625(c)) stored on the at least one memory component (625), the one or more vectors comprising data representative of one or more contiguous segments of one or more neuron weight values (620); retrieve from the at least one memory component (625) the one or more vectors representative of one or more neuron weight values (620) utilizing the one or more vector quantization index values (615); dequantize the retrieved one or more vectors to retrieve underlying one or more neuron weight values (620); and communicate the one or more neuron weight values (620) for processing by the one or more processing components (630) of the neural network environment (600).
  • Claim: 2. The system of claim 1, wherein the one or more vectors are stored in a fast lookup table residing on the at least one memory component.
  • Claim: 3. The system of claim 2, wherein the one or more vectors have an arbitrary length.
  • Claim: 4. The system of claim 3, wherein the computer-readable instructions further cause the at least one processor to retrieve the one or more vectors from one or more rows of the fast lookup table.
  • Claim: 5. The system of claim 4, wherein the vector length of the one or more vectors is selectable for each of the neuron layers of the neural network environment.
  • Claim: 6. The system of claim 5, wherein the computer-readable instructions further cause the at least one processor to perform vector dequantization of one or more neuron weight values for selected one or more of the neuron layers of the neural network environment.
  • Claim: 7. The system of claim 2, wherein the computer-readable instructions further comprising one or more hardware components operative to perform a fast lookup of the vectors stored on the fast look up table.
  • Claim: 8. A computer-implemented method, comprising: receiving one or more initialization parameters from a cooperating controller component (220) of the neural network environment (100), the initialization parameters comprising data representative of the dimensions of the data (340) to be processed by the neural network environment (100) and data representative of one or more vector quantization index values (615), the one or more index values representative of one or more vectors (625(a), 625(b), 625(c)) stored on the at least one memory component (625), the one or more vectors comprising data representative of one or more contiguous segments of one or more neuron weight values (620), the one or more vectors generated by a processor of the neural network environment; retrieving from the at least one memory component (625) the one or more vectors representative of one or more neuron weight values (620) utilizing the one or more vector quantization index values (615), the one or more vectors operatively stored on a fast lookup table; dequantizing the retrieved one or more vectors to retrieve underlying one or more neuron weight values (620); and communicating the one or more neuron weight values (620) for processing by the one or more processing components (630) of the neural network environment (600).
  • Claim: 9. The computer-implemented method of claim 8, further comprising inline dequantization of the one or more retrieved vectors to obtain the one or more neuron weight values by one or more cooperating hardware components of the neural network environment.
  • Claim: 10. The computer-implemented method of claim 8, further comprising utilizing a base index in a cooperating memory component for the one or more generated vectors to generate virtualized one or more fast lookup tables.
  • Claim: 11. The computer-implemented method of claim 8, further comprising generating one or more vectors for one or more neuron layers of the neural network environment.
  • Claim: 12. The computer-implemented method of claim 11, further comprising storing the one or more vectors in one or more rows of a fast lookup table.
  • Claim: 13. The computer-implemented method of claim 12, further comprising generating one or more vectors of arbitrary length.
  • Claim: 14. The computer-implemented method of claim 8, further comprising selecting a vector length for the generation of the one or more vectors for each of the neuron layers of the neural network environment.
  • Claim: 15. The computer-implemented method of claim 8, further comprising storing the one or more generated vectors in a local memory component.
  • Claim: 16. A computer-readable storage medium having computer-executable instructions stored thereupon which, when executed by one or more processors of a computing device, cause the one or more processors of the computing device to: receive one or more initialization parameters from a cooperating controller component (220) of the neural network environment (100), the initialization parameters comprising data representative of the dimensions of the data (340) to be processed by the neural network environment (100) and data representative of one or more vector quantization index values (615), the one or more index values representative of one or more vectors (625(a), 625(b), 625(c)) stored on the at least one memory component (625), the one or more vectors comprising data representative of one or more contiguous segments of one or more neuron weight values (620); retrieve from the at least one memory component (625) the one or more vectors representative of one or more neuron weight values (620) utilizing the one or more vector quantization index values (615); dequantize the retrieved one or more vectors to retrieve underlying one or more neuron weight values (620); and communicate the one or more neuron weight values (620) for processing by the one or more processing components (630) of the neural network environment (600).
  • Claim: 17. The computer-readable storage medium of claim 16, wherein the instructions further cause the one or more processors of the computing device to: store the one or more vectors in one or more fast lookup tables.
  • Claim: 18. The computer-readable storage medium of claim 17, wherein the instructions further cause the one or more processors of the computing device to: select a length of the one or more vectors.
  • Claim: 19. The computer-readable storage medium of claim 16, wherein the instructions further cause the one or more processors of the computing device to: retrieve no vectors for a neuron layer of the neural network environment.
  • Claim: 20. The computer-readable storage medium of claim 16, wherein the instructions further cause the one or more processors of the computing device to: perform inline dequantization of the one or more vectors to retrieve the underlying one or more neuron weight values.
  • Claim: 21. The computer readable medium of claim 16, wherein the memory component cooperates with a physical sensor capable of producing input data comprising audio data, video data, haptic sensory data, and other data for subsequent processing by the one or more cooperating processing units.
  • Claim: 22. The computer readable medium of claim 21, wherein the cooperating processing units electronically cooperate with one or more output physical components operative to receive for human interaction processed input data comprising audio data, video data, haptic sensory data and other data.
  • Current International Class: 06

Klicken Sie ein Format an und speichern Sie dann die Daten oder geben Sie eine Empfänger-Adresse ein und lassen Sie sich per Email zusenden.

oder
oder

Wählen Sie das für Sie passende Zitationsformat und kopieren Sie es dann in die Zwischenablage, lassen es sich per Mail zusenden oder speichern es als PDF-Datei.

oder
oder

Bitte prüfen Sie, ob die Zitation formal korrekt ist, bevor Sie sie in einer Arbeit verwenden. Benutzen Sie gegebenenfalls den "Exportieren"-Dialog, wenn Sie ein Literaturverwaltungsprogramm verwenden und die Zitat-Angaben selbst formatieren wollen.

xs 0 - 576
sm 576 - 768
md 768 - 992
lg 992 - 1200
xl 1200 - 1366
xxl 1366 -