# US 2019/0171926 A1 · Convolutional neural network with sparse and complementary kernels

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0171926 A1
- Document ID: US20190171926A1
- Status: Published application
- Kind: A1
- Application number: 15/829,031
- Filed: 2017-12-01
- Published: 2019-06-06
- Earliest claimed: 2017-12-01
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Richard Chen; Quanfu Fan
- Web page: https://patentgrep.com/patent/US20190171926A1

## CPC

- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/045: Combinations of networks
- G06N 3/0495: Quantised networks; Sparse networks; Compressed networks
- G06N 3/063: using electronic means
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Technologies for providing convolutional neural networks are described. An analysis component determines an initial convolutional layer in a network architecture of a convolutional neural network and one or more subsequent convolutional layers in the network architecture. A replacement component replaces original convolutional kernels in the initial convolutional layer with initial sparse convolutional kernels, and replaces subsequent convolutional kernels in one or more subsequent convolutional layers with complementary sparse convolutional kernels. The complementary sparse kernels have a complementary pattern with respect to sparse kernels of a previous convolutional layer. Analyzing the network architecture and a trained model of a convolutional neural network can determine the original convolutional kernels and replace those kernels with sparse kernels based on similarity and/or weight in an initial layer, with sparse complementary kernels used in subsequent layers.

## Claims

### Claim 1

A system, comprising:
a memory that stores computer executable components;
a processor that executes computer executable components stored in the memory, wherein the computer executable components comprise:
an analysis component that analyzes an initial convolutional layer in a network architecture of a convolutional neural network and one or more subsequent convolutional layers in the network architecture; and
a replacement component that replaces original convolutional kernels in the initial convolutional layer with initial sparse convolutional kernels, and replaces subsequent convolutional kernels in one or more subsequent convolutional layers with complementary sparse convolutional kernels, wherein the complementary sparse convolutional kernels have a complementary pattern with respect to sparse kernels of a previous convolutional layer.

### Claim 2 (depends on 1)

The system of claim 1, wherein the analysis component analyzes a trained model to determine the original sparse kernels.

### Claim 3 (depends on 2)

The system of claim 2, wherein the analysis component determines the sparse kernels based on weight data.

### Claim 4 (depends on 2)

The system of claim 2, wherein the analysis component determines the sparse kernels based on a similarity measure with respect to a most similar sparse pattern inherited from pre-trained weights relative to the original convolutional kernels, and the replacement component uses most similar sparse kernels to replace the original convolutional kernels in the initial convolutional layer.

### Claim 5 (depends on 1)

The system of claim 1, further comprising a training component that uses the initial sparse convolutional kernels and complementary convolutional kernels to train another convolutional neural network model.

### Claim 6 (depends on 1)

The system of claim 1, further comprising a tuning component that uses the initial sparse convolutional kernels and complementary convolutional kernels to tune a convolutional neural network model.

### Claim 7 (depends on 1)

The system of claim 1, wherein relative to a previous convolutional layer of the one or more subsequent convolutional layers, for a kernel, the replacement component replaces the kernel with a complementary sparse kernel relative to the previous convolutional layer or a with another sparse kernel having a most similar sparse pattern.

### Claim 8 (depends on 1)

The system of claim 1, wherein at least one of the sparse kernels comprises a 3×3 kernel size.

### Claim 9 (depends on 1)

The system of claim 1, wherein at least one of the sparse kernels comprises a 5×5 kernel size.

### Claim 10

A computer-implemented method, comprising:
analyzing, by a system operatively coupled to a processor, an initial convolutional layer in a network architecture of a convolutional neural network and one or more subsequent convolutional layers in the network architecture;
replacing, by the system, original convolutional kernels in the initial convolutional layer with initial sparse convolutional kernels; and
replacing, by the system, subsequent convolutional kernels in one or more subsequent convolutional layers with complementary sparse convolutional kernels, wherein the complementary sparse convolutional kernels have a complementary pattern with respect to sparse kernels of a previous convolutional layer.

### Claim 11 (depends on 10)

The computer-implemented method of claim 10, wherein the analyzing further comprises analyzing a trained model to determine the original convolutional kernels.

### Claim 12 (depends on 11)

The computer-implemented method of claim 11, wherein the analyzing determines the sparse kernels based on a similarity measure with respect to a most similar sparse pattern inherited from pre-trained weights relative to the original convolutional kernels.

### Claim 13 (depends on 10)

The computer-implemented method of claim 10, further comprising using, by the system, the complementary sparse convolutional kernels and complementary convolutional kernels to train another convolutional neural network model.

### Claim 14 (depends on 10)

The computer-implemented method of claim 10, further comprising using, by the system, the complementary sparse convolutional kernels and complementary convolutional kernels to tune a convolutional neural network model.

### Claim 15 (depends on 10)

The computer-implemented method of claim 10, wherein the replacing the subsequent convolutional kernels in the one or more subsequent convolutional layers with the complementary sparse convolutional kernels comprises replacing a kernel with a complementary sparse kernel relative to the previous convolutional layer.

### Claim 16

A computer program product facilitating providing efficient convolution neural networks and comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
analyze a network architecture and a trained model of a convolutional neural network to determine original convolutional kernels in an initial convolutional layer and determine one or more subsequent convolutional layers in the network architecture;
replace the original convolutional kernels in the initial convolutional layer with initial sparse convolutional kernels; and
replace subsequent convolutional kernels in one or more subsequent convolutional layers with complementary sparse convolutional kernels, wherein the complementary sparse convolutional kernels have a complementary pattern with respect to sparse kernels of a previous convolutional layer.

### Claim 17 (depends on 16)

The computer program product of claim 16, wherein the program instructions are further executable by the processor to cause the processor to:
determine the original convolutional kernels and the sparse kernels to replace the original convolutional kernels based on weight data.

### Claim 18 (depends on 16)

The computer program product of claim 16, wherein the program instructions are further executable by the processor to cause the processor to:
determine the original convolutional kernels and the sparse kernels to replace the original convolutional kernels based on similarity data.

### Claim 19 (depends on 16)

The computer program product of claim 16, wherein the program instructions are further executable by the processor to cause the processor to:
tune a convolutional neural network model based on the complementary sparse convolutional kernels in the convolutional layers.

### Claim 20 (depends on 16)

The computer program product of claim 16, wherein the program instructions are further executable by the processor to cause the processor to:
train a convolutional neural network model based on the complementary sparse convolutional kernels in the convolutional layers.
