ARES

REQUIREMENTS

To run ARES you will be asked to install the packages listed in requirements.txt. Python 3.10.12 has been used.

EXPERIMENTS

Command for running ARES: python main.py config/configuration.json

In particular, in configuration.json you need to fill the following specifications:

General:

1. dataset: "dataset_name"
2. dataset_extension: ".csv"
3. model_name: "GAE"
4. GPU: 0
5. seed: 42

Data:

1. save_files: 0/1 -- 1 for processing a new dataset from data/raw, 0 for loading a processed dataset from data/processed
2. perc_train: 0.6 -- percentage of data for training
3. perc_val: 0.2 -- percentage of data for validation

Methodology:

1. training: 0/1 -- training the GAE
2. validation: 0/1 -- running the tuning of the hyper-parameters
3. update_hst: 0/1 -- running ARES-Static/ARES-Dynamic
4. phi: "mean"/"minus" -- Eq. 1/2
5. random_start: 0/1 -- HST initialization from training data: 0 -> last window, 1 -> uniform sampling.

Training:

1. in_channels: 32
2. hidden_channels: 16
3. out_channels: 8
4. num_layers: 3
5. epochs: 10000
6. learning_rate: 0.001
7. patience: 1000
8. plot_title: "plot_loss/plot_loss_GAE_dataset_name"
9. model_save_path: "checkpoints/model_GAE_dataset_name"

Test:

1. n_trees: 8
2. height: 6
3. window_size: 64
4. thresholds: "0.859, 0.739, 0.876, 0.854, 0.840, 0.865, 0.901, 0.855, 0.875",
5. weights: "1.0,0.0,0.0" -- w_1, w_2, and w_3 from Eq 4
6. window_already_seen: 64 -- Number of steps before resetting the cache memory

Note:

1. Configuration files for the dataset used are in the "config" folder.
2. data/processed/IDS2018 and data/processed/DARPA need to be uncompressed due to the size dimension limit on github.

Cite us

If you use ARES in your research, please cite our KDD 2026 paper:

@article{mungari2025ares,
  title={ARES: Anomaly Recognition Model For Edge Streams},
  author={Mungari, Simone and Bifet, Albert and Manco, Giuseppe and Pfahringer, Bernhard},
  journal={arXiv preprint arXiv:2511.22078},
  year={2025}
}