mohamad-ansarimehr/F-ACVAE: Fourth release of F-ACVAE implementation

Fourth release of the F-ACVAE implementation. This release provides the updated source code for F-ACVAE: A Federated Adaptive Conditional Variational Auto-Encoder for Privacy-Preserving Intrusion Detection in IoT Networks. Key updates and features: Custom Constrained Momentum Gaussian Aggregation (CMGA) strategy implemented in the Flower framework Selective aggregation with private client encoders and shared decoder/class-conditioning parameters Client-local Random Forest training and evaluation without uploading latent representations Class-wise Gaussian prior coordination using client-reported latent statistics Non-IID federated evaluation on the N-BaIoT dataset using Dirichlet-based client partitioning Cross-dataset validation on UNSW-NB15 and CIC-IDS2017 Ablation experiments for CMGA, selective aggregation, adaptive conditioning, and the downstream classifier Repeated-run stability evaluation across multiple training seeds Approximately 50% reduction in communicated trainable parameters compared with full-model aggregation PyTorch and Flower implementation with reproducible experiment settings This release also clarifies the privacy scope of the framework: raw data, private encoders, and the primary downstream classifier remain local, while shared model parameters and class-wise Gaussian statistics are communicated during federated training. The implementation does not claim a formal differential-privacy guarantee. For details, see the accompanying paper (under review/submission): "F-ACVAE: A Federated Adaptive Conditional Variational Auto-Encoder for Privacy-Preserving Intrusion Detection in IoT Networks" by Mohammad Ansarimehr et al. README.md includes installation instructions, dependencies, experiment configuration, and steps for reproducing the reported results.