Table 7 Baseline architecture and hyperparameter configurations.
From: Multimodal fusion based few-shot network intrusion detection system
Model architecture (stacking times/fusion method) | Hyperparameter | Value |
|---|---|---|
Upper Layers G-Model (2) | Padding | 1 |
Lower Layers G-Model (5) | Dropout | 0.1 |
S-Model-Transf. (6) | Heads | 4 |
Dim | 32 | |
Depth | 6 | |
attn_dropout | 0.1 | |
ff_dropout | 0.1 | |
S-Model-MLP | mlp_mults | (4, 2) |
Self-sufficient model (LF) | lr | 0.001 |
Dropout | 0.1 | |
Optimizer | adam | |
Loss function | Cross entropy | |
Transfer-enhanced model (LF) | lr | 0.0001 |
Dropout | 0.3 | |
Common settings | Batch size | 32 |
Epochs | 200 | |
Fusion dim | 128 | |
\(k_{\text {train}}\) | [5, 10, 15] | |
\(k_{\text {test}}\) | 30 | |
\(k_{\text {source}}\) | 100 | |
\(n_{\text {way}}\) | 5 |
