Graph Neural Network Layer Count Determination via Reinforcement Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Graph neural networks fail to consider heterogeneous characteristics of users and items, leading to suboptimal performance in recommendation systems, as they apply a uniform number of layers regardless of node characteristics in the graph structure.

Innovation Solution

An apparatus and method using reinforcement learning models to determine the optimal number of layers for each node in the graph neural network based on user and item-specific characteristics, by setting rewards for purchase information and adapting the number of layers for embedding extraction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a uniform number of layers is applied to all nodes in the graph neural network, then the device complexity is reduced and ease of operation is improved, but the adaptability to heterogeneous characteristics of different nodes deteriorates

Engineering Contradiction:

Improveadaptability to heterogeneous characteristicsVSAvoidcomplexity of determining layer numbers

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent makes the number of layers dynamic by using a reinforcement learning model to adaptively determine the optimal number of layers for each node based on its heterogeneous characteristics. Instead of a fixed uniform structure, the system dynamically adjusts layer configuration per node to maximize performance while considering individual node properties.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent applies local quality by allowing different nodes in the graph structure to have different numbers of layers tailored to their specific characteristics. Each node receives a customized layer configuration determined by the reinforcement learning model based on local node features, rather than applying a global uniform structure.

Inventive Principle:

Principle #3Local quality

2Adaptability or versatility

If the number of layers is increased to capture heterogeneous characteristics, then the adaptability improves, but the training time and computational cost increase

Engineering Contradiction:

Improveability to consider heterogeneous characteristicsVSAvoidtraining time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent changes the parameter of number of layers from a fixed value to a dynamically determined value optimized for each node. The reinforcement learning model learns the optimal layer parameter for each node based on its characteristics, achieving adaptability without uniformly increasing layers across all nodes, thus avoiding unnecessary computational overhead.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent applies partial action by determining the number of layers on demand for each node rather than pre-configuring a maximum depth for all nodes. The reinforcement learning model determines the exact number of layers needed for each node, avoiding excessive computation on nodes that require fewer layers.

Inventive Principle:

Principle #16Partial or excessive action

3Productivity

If reinforcement learning model is used to determine optimal layers for each node, then the performance improves, but the device complexity and implementation difficulty increase

Engineering Contradiction:

Improveperformance of recommendation systemVSAvoidcomplexity of reinforcement learning integration

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent applies universality by using a single reinforcement learning model framework that can handle multiple nodes with different characteristics. The same RL model architecture serves all nodes in the graph, determining optimal layer numbers for each based on their specific features, thus avoiding the need for separate specialized systems for different node types.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The reinforcement learning model performs self-service by automatically determining the optimal number of layers for each node without manual configuration or intervention. The system trains the RL model to autonomously make decisions about layer configuration based on node characteristics and performance feedback.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20240119299A1Apparatus for determining the number of layers of graph neural network by using reinforcement learning model, method for determining the number of layers of graph neural network by using reinforcement learning model, and recording medium storing instructions to perform method for determining the number of layers of graph neural network by using reinforcement learning model

Publication Date: 2024.04.11 RES & BUSINESS FOUND SUNGKYUNKWAN UNIV

AI summary

In accordance with an aspect of the present disclosure, there is provided an apparatus for determining a number of layers may comprise a data manager configured to obtain a graph structure including information between nodes; a first controller configured to control a reinforcement learning model designed to set the graph structure as an environment of reinforcement learning, set any one of nodes included in the graph structure as a state of reinforcement learning, and set a number of branches causing a highest expected value for a reward of reinforcement learning from any node as an action of reinforcement learning; a storage configured to store the number of branches determined by the reinforcement learning model as an action for a predetermined node input to the reinforcement learning model; and a second controller configured to apply the same number of layers for extracting an embedding of the predetermined node as the stored number of branches in a graph neural network designed based on the graph structure.