Deep Learning–Enabled Multimodal Continuous Authentication
Abstract
Traditional authentication mechanisms rely on static credentials and session-based validation, which are insufficient against modern threats such as session hijacking and credential misuse. Continuous authentication addresses this gap by validating user identity throughout an active session; however, practical deployment in enterprise environments introduces challenges related to accuracy, usability, and system complexity.
This paper presents a deep learning–enabled multimodal continuous authentication system that combines behavioral and contextual signals to continuously assess user identity. The proposed architecture applies deep learning–based feature extraction and fusion to improve authentication confidence while minimizing user friction. The system is designed with enterprise constraints in mind, including latency, privacy, and scalability.
Experimental evaluation demonstrates that multimodal fusion improves authentication robustness compared to single-modality approaches. Beyond model performance, this work discusses architectural trade-offs and deployment considerations relevant to modern identity and access management platforms.
Problem Context
Session-based authentication provides only point-in-time assurance and fails to detect identity compromise after login. Enterprise systems require continuous assurance without degrading user experience or operational stability.
System Overview
The proposed system continuously evaluates user identity by combining multiple behavioral and contextual signals. Each modality is processed independently and fused using a deep learning model to generate a continuous confidence score.
Deep Learning Approach
The system applies deep learning techniques for feature extraction and multimodal fusion. Model selection prioritizes robustness, adaptability, and suitability for real-time authentication scenarios.
Evaluation Summary
Experimental results indicate improved accuracy and stability when using multimodal fusion compared to single-modality authentication. Observations focus on practical performance trends rather than isolated benchmark metrics.
Key Contributions
- Designed a scalable multimodal continuous authentication architecture
- Applied deep learning–based fusion for continuous identity assurance
- Identified deployment considerations for enterprise identity systems
Full Paper
📄 Full technical paper (PDF)
(To be added)