Statement Recognition of Access Control Policies in IoT Networks

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Access Control Policies (ACPs) are critical for securing Internet of Things (IoT) networks.
• Recognizing ACPs from natural language documents is challenging due to limited labeled data for individual clients.
• Existing methods often overlook the data scarcity issue, hindering model accuracy.

Purpose of the Study:

• To propose a novel model, FL-Bert-BiLSTM, for accurate and privacy-preserving access control policy recognition in IoT networks.
• To address the challenge of limited labeled data in federated environments.
• To leverage federated learning and pre-trained word embeddings for enhanced ACP recognition.

Main Methods:

• Developed FL-Bert-BiLSTM, a hybrid model combining federated learning with BiLSTM and pre-trained word embeddings.
• Implemented a distributed training approach on IoT devices to overcome data scarcity.
• Utilized pre-trained word embeddings to capture semantic information from policy documents.

Main Results:

• The FL-Bert-BiLSTM model significantly improved accuracy and generalization capability in ACP recognition.
• The federated learning approach effectively mitigated the scarcity of labeled data.
• The model demonstrated robust data privacy preservation during distributed training.

Conclusions:

• FL-Bert-BiLSTM offers a viable solution for secure and efficient access control policy recognition in IoT networks.
• The combination of federated learning and pre-trained embeddings enhances model performance and data privacy.
• The proposed model is well-suited for real-time, distributed applications in IoT environments.