Fig 1.
A methodology for Zero-Day Vulnerability Identification.
Table 1.
DQN agent hyperparameters selection at the time of training.
Fig 2.
DQN agent training log with 1000 episodes and episode reward.
Table 2.
DQN agent training statistics after completing 1000 episodes.
Fig 3.
Training Performance of DQN Agent with 1000 Episodes and Episode Steps.
Fig 4.
DQN Agent Training Progress: Episode Rewards over 1000 Episodes.
Fig 5.
Tracking the performance of a DQN agent over 1000 training episodes.
Fig 6.
Measuring the Performance of DQN Agent in 1000 Training Episodes and average steps.
Fig 7.
Training coverage of the agent.
Fig 8.
DQN agent Training scatter plot.
Table 3.
Training output sample data are taken from the DQN agent.
Fig 9.
DQN Agent Forecasting Progress: Episode Rewards after forecasted 1000 Episodes.
Fig 10.
DQN Agent with 1000 training and 1000 forecasted Episode Steps Performance.
Fig 11.
Performance of the DQN Agent: Episode Rewards after Predicted 1000 Episodes.
Fig 12.
Performance of DQN agent over 1000 training episodes and 1000 forecasted episodes.
Fig 13.
Measure Performance of DQN Agent Predicting Average Steps in 1000 Episodes.
Table 4.
DQN agent Forecasting statistics after completing 1000 episodes.
Table 5.
Comparison of Reinforcement Learning Approaches in Cybersecurity Research.
Fig 14.
Visualization of the Relationships between Different Metrics in a Normalized and Scaled Forecast Statistics.
Fig 15.
Comparison of Multiple Runs of Same Algorithm on the Same Dataset (Kaggle-RL).
Fig 16.
Comparison of Multiple Runs of Same Algorithm on the Same Dataset (NSL-KDD).
Fig 17.
Comparison of Multiple Runs of Same Algorithm on the Same Dataset (UNSW-NB15).
Table 6.
Comparison of Deep Learning (DL), Traditional ML and Proposed RL Methodology (DQN).