AI Model Training with Interval Sampling for NPC Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing game AI models for non-player characters (NPCs) in open world games face challenges in training due to the vast decision-making space and real-time requirements, necessitating improved generalization and efficiency in model training.
Innovation Solution
An AI model training method involving reinforcement learning with random sampling and numerical interval determination is employed to generate a training character set, ensuring extensive sample data and improved generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional training methods are used for game AI models, then manual maintenance operations are required, but training efficiency is low and generalization is limited
Solution Approach 1:
The AI model performs self-training by automatically generating training data through random sampling of attribute data within determined numerical intervals. The model uses its own outputs to create training character sets without requiring manual data preparation or maintenance operations, enabling autonomous improvement of training efficiency
Solution Approach 2:
The system pre-determines numerical intervals for attribute data before training begins. By establishing these intervals in advance and using them for random sampling during training, the system prepares the training environment beforehand, reducing the need for manual adjustments during the training process and improving overall efficiency
2Adaptability or versatility
If the decision-making space for NPCs is expanded to provide richer strategic variations, then the complexity of training increases
Solution Approach 1:
The attribute data is segmented into multiple numerical intervals, and the training process is divided into generating training character sets from these intervals. By segmenting the vast decision-making space into manageable intervals and generating targeted training data from each, the system reduces the perceived complexity while maintaining comprehensive coverage of strategic variations
Solution Approach 2:
The system changes the parameters of training data generation by using random sampling within determined numerical intervals. This parameter-based approach allows the model to explore different strategic variations systematically by adjusting the intervals and sampling processes, making the training of complex decision-making spaces more manageable
3Quantity of substance
If real-time decision-making requirements are met, then the training data must be extensive and diverse, but traditional sampling methods generate limited sample data
Solution Approach 1:
The training process continuously generates training character sets by repeatedly sampling attribute data within the determined numerical intervals. This continuous generation process ensures that extensive and diverse sample data are produced efficiently, meeting real-time decision-making requirements without requiring lengthy training periods
Solution Approach 2:
The system dynamically adjusts the training data generation process by using random sampling within flexible numerical intervals. This dynamic approach allows the training system to adaptively generate diverse sample data as needed, ensuring both quantity and quality of training data are met within acceptable time frames
Data Source
AI summary
Provided are an artificial intelligence model training method and apparatus, a device, and a storage medium, which are used for rapid training to obtain an AI model while improving the generalization of the AI model. The method includes: obtaining an initial artificial intelligence (AI) model and a first numerical interval, the first numerical interval being a numerical value range of a plurality of pieces of attribute data corresponding to an opponent character of the initial AI model; invoking the initial AI model to determine a second numerical interval from the first numerical interval; performing random sampling on the plurality of pieces of attribute data within the second numerical interval to generate a training character set; and performing reinforcement learning training on the initial AI model by using the training character set, to obtain a target AI model.