# US 11,443,163 B2 · Method and system for executing neural network > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,163 B2 - Document ID: US11443163B2 - Status: Granted - Kind: B2 - Application number: 16/600,299 - Filed: 2019-10-11 - Published: 2021-04-15 - Granted: 2022-09-13 - Earliest claimed: 2019-10-11 - Assignees: Alibaba Group Holding Limited - Applicants: Alibaba Group Holding Limited - Inventors: Minghai Qin; Chunsheng Liu - Web page: https://patentgrep.com/patent/US11443163B2 ## CPC - G06N 3/04: Architecture, e.g. interconnection topology - G06N 3/048: Activation functions - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/045: Combinations of networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract The present disclosure provides methods and systems for executing a neural network. The method includes: receiving a plurality of input vectors of input data; generating, among the plurality of input vectors, an estimation value associated with a subset of an input vector based on a weight vector of the activation function; determining whether the estimation value associated with the subset of the input vector satisfies a threshold condition; and determining an output of the activation function based on the estimation value. ## Claims ### Claim 1 A computer-implemented method for executing an activation function of a neural network, comprising: receiving a plurality of input vectors of input data; generating, among the plurality of input vectors, an estimation value associated with a subset of an input vector based on a weight vector of the activation function; determining whether the estimation value associated with the subset of the input vector satisfies a threshold condition; and determining an output of the activation function based on the estimation value. ### Claim 2 (depends on 1) The method according to claim 1, wherein determining the output of the activation function based on the estimation value further comprises: in response to the estimation value not satisfying the threshold condition, setting the output to zero; or in response to the estimation value satisfying the threshold condition, determining the output by performing a dot production on the input vector and the weight vector. ### Claim 3 (depends on 1) The method according to claim 1, wherein the weight vector is comprised of a first subset of weight elements and a second subset of weight elements, the input vector is comprised of a first subset of input elements and a second subset of input elements, and generating the estimation value associated with the subset of the input vector based on the weight vector of the activation function further comprises: determining the estimation value based on the first subset of weight elements and the first subset of input elements, wherein the first subset of input elements corresponds to the first subset of weight elements. ### Claim 4 (depends on 3) The method according to claim 3, wherein determining the estimation value based on the first subset of weight elements and the first subset of input elements further comprises: performing a dot production on the first subset of weight elements and the first subset of input elements to generate a first dot product as the estimation value. ### Claim 5 (depends on 3) The method according to claim 3, wherein the first subset of weight elements comprises a number of largest absolute weight values of the weight vector. ### Claim 6 (depends on 3) The method according to claim 3, wherein a size of the first subset of weight elements is equal to a size of the first subset of input elements. ### Claim 7 (depends on 4) The method according to claim 4, wherein determining the output by performing the dot production on the input vector and the weight vector comprises: performing the dot production on the second subset of weight elements and the second subset of input elements to generate a second dot product; and generate the output by adding the estimation value and the second dot product. ### Claim 8 (depends on 1) The method according to claim 1, further comprising: determining estimation values for each of the plurality of input vectors; determining dot products of the plurality of input vectors and the weight vector based on the estimation values; and generating a result of the neural network based on the dot products. ### Claim 9 (depends on 1) The method according to claim 1, wherein the threshold condition comprises the estimation value associated with the input vector being greater than or equal to a threshold. ### Claim 10 (depends on 9) The method according to claim 9, wherein the threshold of the threshold condition is determined based on a type of the neural network. ### Claim 11 A system for executing a neural network, comprising: a memory storing a set of instructions; and at least one processor configured to execute the set of instructions to cause the system to perform: receiving a plurality of input vectors of input data; generating, among the plurality of input vectors, an estimation value associated with a subset of an input vector based on a weight vector of the activation function; determining whether the estimation value associated with the subset of the input vector satisfies a threshold condition; and determining an output of the activation function based on the estimation value. ### Claim 12 (depends on 11) The system according to claim 11, wherein determining the output of the activation function based on the estimation value further comprises: in response to the estimation value not satisfying the threshold condition, setting the output to zero; or in response to the estimation value satisfying the threshold condition, determining the output by performing a dot production on the input vector and the weight vector. ### Claim 13 (depends on 11) The system according to claim 11, wherein the weight vector is comprised of a first subset of weight elements and a second subset of weight elements, the input vector is comprised of a first subset of input elements and a second subset of input elements, and generating the estimation value associated with the subset of the input vector based on the weight vector of the activation function further comprises: determining the estimation value based on the first subset of weight elements and the first subset of input elements, wherein the first subset of input elements corresponds to the first subset of weight elements. ### Claim 14 (depends on 13) The system according to claim 13, wherein determining the estimation value based on the first subset of weight elements and the first subset of input elements further comprises: performing a dot production on the first subset of weight elements and the first subset of input elements to generate a first dot product as the estimation value. ### Claim 15 (depends on 13) The system according to claim 13, wherein the first subset of weight elements comprises a number of largest absolute weight values of the weight vector. ### Claim 16 (depends on 13) The system according to claim 13, wherein a size of the first subset of weight elements is equal to a size of the first subset of input elements. ### Claim 17 (depends on 14) The system according to claim 14, wherein determining the output by performing the dot production on the input vector and the weight vector comprises: performing the dot production on the second subset of weight elements and the second subset of input elements to generate a second dot product; and generate the output by adding the estimation value and the second dot product. ### Claim 18 (depends on 11) The system according to claim 11, wherein the set of instructions is further executed to cause the system to perform: determining estimation values for each of the plurality of input vectors; determining dot products of the plurality of input vectors and the weight vector based on the estimation values; and generating a result of the neural network based on the dot products. ### Claim 19 (depends on 11) The system according to claim 11, wherein the threshold condition comprises the estimation value associated with the input vector being greater than or equal to a threshold. ### Claim 20 A non-transitory computer readable medium that stores a set of instructions that is executable by at least one processor of a computer system to cause the computer system to perform a method for executing an activation function of a neural network, the method comprising: receiving a plurality of input vectors of input data; generating, among the plurality of input vectors, an estimation value associated with a subset of an input vector based on a weight vector of the activation function; determining whether the estimation value associated with the subset of the input vector satisfies a threshold condition; and determining an output of the activation function based on the estimation value.