# US 12,536,441 B2 · Method and system for changing structure of deep learning model based on change in input resolution

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

- Number: US 12,536,441 B2
- Document ID: US12536441B2
- Status: Granted
- Kind: B2
- Application number: 17/819,261
- Filed: 2022-08-11
- Published: 2023-02-16
- Granted: 2026-01-27
- Earliest claimed: 2021-08-12
- Assignees: Nota, Inc.
- Applicants: Nota, Inc.
- Inventors: Byeongman Lee; Jae Woong Yun; Minsu Kim; Tae-Ho Kim
- Web page: https://patentgrep.com/patent/US12536441B2

## CPC

- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Disclosed are a method and system for changing a structure of a deep learning model based on a change in resolution of input data. The method of changing a structure of a deep learning model may include generating, by the at least one processor, a plurality of input data having different resolution by performing various resolution changes on input data having given resolution, performing, by the at least one processor, inference on each of the plurality of generated input data through a deep learning model, checking, by the at least one processor, the size of a feature map output by each of layers included in the deep learning model while the inference is performed, and changing, by the at least one processor, the structure of at least one of the layers based on the checked size of the feature map.

## Claims

### Claim 1

A method of changing a structure of a deep learning model, which is performed by a computer device comprising at least one processor, the method comprising:
generating, by the at least one processor, a plurality of input data having different resolution, respectively, by performing various resolution changes on input data having given resolution, the different resolution being smaller than the given resolution;
performing, by the at least one processor, inference on each of the plurality of generated input data through the deep learning model;
identifying, by the at least one processor, a layer that outputs a feature map having a size equal to or less than a predetermined size among a plurality of layers included in the deep learning model while the inference is performed based on the plurality of generated input data; and
removing, by the at least one processor, the identified layer from the deep learning model or modifying a structure of the identified layer such that the identified layer outputs a feature map larger than the predetermined size to change the structure of the deep learning model.

### Claim 2 (depends on 1)

The method of claim 1, wherein the predetermined size of the feature map output by the identified layer is 1×1.

### Claim 3 (depends on 1)

The method of claim 1, wherein the deep learning model is a model trained by learning data having the given resolution.

### Claim 4 (depends on 1)

The method of claim 1, wherein the deep learning model comprises a convolutional neural network (CNN).

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
generating output data using the deep learning model with the changed structure based on the input data.

### Claim 6 (depends on 1)

The method of claim 1, wherein the modifying of the structure of the identified layer comprises adjusting a parameter of the identified layer associated with at least one of a stride or a padding of the identified layer.

### Claim 7 (depends on 1)

A non-transitory computer-readable recording medium on which a program for executing the method according to claim 1 in a computer device is recorded.

### Claim 8

A computer device comprising at least one processor implemented to execute an instruction readable by a computer device,
wherein the at least one processor is configured to:
generate a plurality of input data having different resolution, respectively, by performing various resolution changes on input data having given resolution, the different resolution being smaller than the given resolution,
perform inference on each of the plurality of generated input data through a deep learning model,
identify a layer that outputs a feature map having a size equal to or less than a predetermined size among a plurality of layers included in the deep learning model while the inference is performed based on the plurality of generated input data, and
remove the identified layer from the deep learning model or modify a structure of the identified layer such that the identified layer outputs a feature map larger than the predetermined size to change a structure of the deep learning model.

### Claim 9 (depends on 8)

The computer device of claim 8, wherein the predetermined size of the feature map output by the identified layer is 1×1.

### Claim 10 (depends on 8)

The computer device of claim 8, wherein the deep learning model is a model trained by learning data having given resolution.

### Claim 11 (depends on 8)

The computer device of claim 8, wherein the at least one processor is configured to modify the structure of the identified layer by adjusting a parameter of the identified layer associated with at least one of a stride or a padding of the identified layer.