Parallel MLP Image Classification for Faster Patch Processing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image classification technologies face challenges in achieving both speed and accuracy, particularly when using multi-layer perceptron (MLP) models, as they often require sequential processing of image patches which limits efficiency.

Innovation Solution

Implementing a parallel structure of multiple MLP layers that operate concurrently, where each layer includes multiple MLPs, allowing for parallel processing of image patches to learn local and relational characteristics, and combining their results using attention weights to maintain consistent data dimensions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If sequential processing of image patches is used in MLP models, then processing simplicity is maintained, but classification speed deteriorates

Engineering Contradiction:

Improveclassification speedVSAvoidmodel structure complexity

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The image is divided into multiple patches, and each patch is processed by separate MLP branches in parallel. This segmentation allows simultaneous processing of different image regions, significantly improving classification speed while maintaining manageable model complexity through modular architecture

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The model transitions from sequential single-branch processing to parallel multi-branch processing by adding a dimensional aspect. Multiple MLP branches operate concurrently on different image patches, transforming the processing paradigm from one-dimensional sequential to multi-dimensional parallel operations

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Productivity

If parallel MLP layers are used to process image patches simultaneously, then processing speed improves, but model complexity increases

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidnumber of MLP layers

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

Multiple MLP branches that process different image patches are merged into a unified architecture with shared components. The branches combine their outputs through concatenation or addition, achieving parallel processing efficiency while reducing overall model complexity through parameter sharing and integrated structure

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The MLP branches are designed with universal, reusable components that can process different image patches using the same learned parameters. This multi-functionality allows the model to maintain high processing efficiency through parallelization while avoiding the complexity increase that would result from entirely separate processing paths

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12423951B2Server for classifying image and operating method thereof

Publication Date: 2025.09.23 SAMSUNG ELECTRONICS CO LTD

AI summary

A server for classifying an image and a method of operating the server are provided. The method includes obtaining classification probability values of the image, by applying the image to an image classification model including a plurality of parallel multi-layer perceptron (MLP) layers, and classifying the image, based on the classification probability values, wherein each of the plurality of parallel MLP layers includes a first MLP and a second MLP, an operation using the first MLP and an operation using the second MLP are performed in parallel, and dimensions of data before and after an operation of each of the plurality of parallel MLP layers are same by combining an operation result of the first MLP with an operation result of the second MLP.