公开(公告)号：US11604987B2

公开(公告)日：2023-03-14

申请号：US16826472

申请日：2020-03-23

申请人： QUALCOMM Incorporated

发明人： Marinus Willem Van Baalen ,  Tijmen Pieter Frederik Blankevoort ,  Markus Nagel

IPC分类号： G06N3/08 ,  G06N3/04

摘要： Various embodiments include methods and neural network computing devices implementing the methods, for generating an approximation neural network. Various embodiments may include performing approximation operations on a weights tensor associated with a layer of a neural network to generate an approximation weights tensor, determining an expected output error of the layer in the neural network due to the approximation weights tensor, subtracting the expected output error from a bias parameter of the layer to determine an adjusted bias parameter and substituting the adjusted bias parameter for the bias parameter in the layer. Such operations may be performed for one or more layers in a neural network to produce an approximation version of the neural network for execution on a resource limited processor.

公开(公告)号：US12131258B2

公开(公告)日：2024-10-29

申请号：US17030315

申请日：2020-09-23

申请人： QUALCOMM Incorporated

发明人： Yadong Lu ,  Ying Wang ,  Tijmen Pieter Frederik Blankevoort ,  Christos Louizos ,  Matthias Reisser ,  Jilei Hou

IPC分类号： G06N3/082 ,  G06N3/04 ,  G06N3/063

CPC分类号： G06N3/082 ,  G06N3/04 ,  G06N3/063

摘要： A method for compressing a deep neural network includes determining a pruning ratio for a channel and a mixed-precision quantization bit-width based on an operational budget of a device implementing the deep neural network. The method further includes quantizing a weight parameter of the deep neural network and/or an activation parameter of the deep neural network based on the quantization bit-width. The method also includes pruning the channel of the deep neural network based on the pruning ratio.

公开(公告)号：US11562208B2

公开(公告)日：2023-01-24

申请号：US16413535

申请日：2019-05-15

申请人： QUALCOMM Incorporated

发明人： Christos Louizos ,  Matthias Reisser ,  Tijmen Pieter Frederik Blankevoort ,  Max Welling

IPC分类号： G06N3/04 ,  G06N3/08 ,  G06F17/18

摘要： A method for quantizing a neural network includes modeling noise of parameters of the neural network. The method also includes assigning grid values to each realization of the parameters according to a concrete distribution that depends on a local fixed-point quantization grid and the modeled noise and. The method further includes computing a fixed-point value representing parameters of a hard fixed-point quantized neural network.

公开(公告)号：US11704571B2

公开(公告)日：2023-07-18

申请号：US17067233

申请日：2020-10-09

申请人： QUALCOMM Incorporated

发明人： Kambiz Azarian Yazdi ,  Tijmen Pieter Frederik Blankevoort ,  Jin Won Lee ,  Yash Sanjay Bhalgat

IPC分类号： G06N3/08 ,  G06N3/082 ,  G06N3/04

CPC分类号： G06N3/082 ,  G06N3/04

摘要： A method for pruning weights of an artificial neural network based on a learned threshold includes determining a pruning threshold for pruning a first set of pre-trained weights of multiple pre-trained weights based on a function of a classification loss and a regularization loss. Weights are pruned from the first set of pre-trained weights when a first value of the weight is less than the pruning threshold. A second set of pre-trained weights of the multiple pre-trained weights is fine-tuned or adjusted in response to a second value of each pre-trained weight in the second set of pre-trained weights being greater than the pruning threshold.

公开(公告)号：US20200372361A1

公开(公告)日：2020-11-26

申请号：US16419509

申请日：2019-05-22

申请人： QUALCOMM Incorporated

发明人： Babak Ehteshami Bejnordi ,  Tijmen Pieter Frederik Blankevoort ,  Max Welling

IPC分类号： G06N3/08 ,  G06N20/00

摘要： A computing device may be equipped with a generalized framework for accomplishing conditional computation or gating in a neural network. The computing device may receive input in a neural network layer that includes two or more filters. The computing device may intelligently determine whether the two or more filters are relevant to the received input. The computing device may deactivate filters that are determined not to be relevant to the received input (or activate filters that are determined to be relevant to the received input), and apply the received input to active filters in the layer to generate an activation.