A Human-Machine Agent Based on Active Reinforcement Learning for Target Classification in Wargame.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Military Technology

Background:

• Modern warfare demands high-accuracy, low-cost target classification for threat assessment.
• Existing methods struggle to balance human expertise with machine learning efficiency.

Purpose of the Study:

• To propose a human-machine agent for target classification using active reinforcement learning (TCARL_H-M).
• To determine optimal human guidance integration and autonomous classification strategies.
• To evaluate the model's performance against various benchmarks.

Main Methods:

• Developed TCARL_H-M with two human guidance modes (low-value cues vs. high-value labels).
• Introduced machine-only (TCARL_M) and human-only (TCARL_H) models for comparison.
• Utilized wargame simulation data for performance evaluation.

Main Results:

• TCARL_H-M significantly reduced labor costs.
• Achieved competitive classification accuracy compared to TCARL_M, TCARL_H, LSTM, QBC, and Uncertainty sampling.
• Demonstrated effective target prediction and classification capabilities.

Conclusions:

• TCARL_H-M offers a cost-effective and accurate solution for target classification.
• The hybrid human-machine approach enhances classification performance in simulated warfare scenarios.