Table 1.
Summary of meta-learning related works.
Table 2.
Summary of self-attention related works.
Table 3.
Summary of few-shot IoT intrusion detection works.
Fig 1.
The overall architecture of MACML-IDS.
Fig 2.
The overall architecture of the MAC module.
Fig 3.
Training process of the MAC module.
Table 4.
Data types in CICIoT2023 dataset.
Table 5.
Data types in CICIDS2018 dataset.
Table 6.
Hyperparameter settings for MACML-IDS.
Fig 4.
Accuracy trends during training.
Fig 5.
Detection rate trends during training.
Fig 6.
MACML-IDS performance evaluation on CICIoT2023 dataset with different training sample sizes.
Fig 7.
MACML-IDS performance evaluation on CICIDS2018 dataset with different training sample sizes.
Table 7.
F1-score for different attack types with different training sample sizes.
Fig 8.
Data distribution of the results of testing all attack types (CICIoT2023).
Fig 9.
Data distribution of the results of testing all attack types (CICIDS2018).
Table 8.
Cross-domain detection results on CICIoT2023.
Table 9.
Cross-domain detection results on CICIDS2018.
Fig 10.
Comparison of same-domain and cross-domain experimental results on CICIoT2023.
Fig 11.
Comparison of same-domain and cross-domain experimental results on CICIDS2018.
Fig 12.
Data distribution of the cross-domain experimental results on CICIoT2023.
Fig 13.
Data distribution of the cross-domain experimental results on CICIDS2018.
Table 10.
Comparison between MACML-IDS and related work.
Fig 14.
Visualization of the CICIoT2023 dataset.
Fig 15.
Visualization of the CICIDS2018 dataset.
Fig 16.
Heatmaps of average confusion matrices in 1, 5 and 10-shot scenarios.
Fig 17.
Multiclass detection results in 1, 5 and 10-shot scenarios.