LINC is designed to enable in-network classification tasks within low-resource network environments and provides an efficient strategy for incremental model updates. See our ICNP 24 paper for more details!
Before you dive in, make sure to install all the required dependencies.
pip install -r requirements.txt
In this section, we'll guide you through training a Neural Network model and converting it into a set of rules.
We utilize the ISCX dataset for training the Neural Network model. You can download the preprocessed dataset and place the iscx.pkl file in the root directory of the project.
The features used in the ISCX dataset are as follows:
{"protocol": 8, "ip_ihl": 4, "ip_tos": 8, "ip_flags": 8, "ip_ttl": 8, "tcp_dataofs": 4, "tcp_flags": 8, "tcp_window": 16, "udp_len": 16, "length": 16}
Where these integers indicate how many bits used to represent the features. Each of these features is converted into several different bit features, and all the bit features are concatenated to a 96-dimensional vector as input for LINC.
Run the following command to train the Neural Network model and convert it into a set of rules:
python train_nn.py
During execution, the log will display detailed information about the training and conversion process, including model accuracy, the number of derived rules, and conversion time. The converted rules will be saved in the checkpoints/iscx/rules.txt file.
After model training and conversion, we counted the number of times each bit feature was selected by aggregating across all the learned rules.
In this section, we demonstrate how to incrementally update LINC rules with just a few labeled samples. Consistent with our paper, we first train a Neural Network model using three categories of traffic types (Chat, File Transfer, and Email).
python train_nn_c3.py
All results will be stored in the checkpoints/iscx_c3/ directory.
Next, we proceed to update the model using only 63 labeled samples from the Streaming category:
python update_streaming.py
The results of the update will be saved in the checkpoints/iscx_c4/ directory.