Smartphone Impostor Detection via Local RNN and SID
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Smartphone theft poses a significant threat as impostors can bypass initial security measures, compromising user confidentiality and privacy, and existing solutions require users to share sensitive behavioral data for training, raising privacy concerns.
Innovation Solution
A recurrent neural network (RNN)-based deep learning algorithm using sensor data from legitimate users to learn normal behavior, with a minimalist hardware module called the smartphone impostor detector (SID) for real-time detection, which protects user privacy by not exposing data outside the device and supports both anomaly detection and machine learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor data is collected and analyzed to detect impostors, then detection accuracy is improved, but user privacy is compromised due to data exposure
Solution Approach 1:
The patent introduces an intermediary processing layer that transforms raw sensor data into behavioral features and templates. This intermediary representation preserves the essential information needed for detection while removing personally identifiable information, thus mediating between detection accuracy and privacy protection
Solution Approach 2:
The system creates behavioral templates as copies of user behavior patterns rather than storing actual sensor data. These templates capture the essence of user behavior (typing rhythm, swipe patterns, gait) without containing sensitive personal information, enabling detection while preserving privacy
2Measurement precision
If deep learning algorithms are used for impostor detection, then detection accuracy is improved, but hardware cost and energy consumption increase
Solution Approach 1:
The patent extracts and implements only the essential deep learning components needed for impostor detection, rather than deploying full-scale deep learning systems. By taking out only the necessary computational elements and optimizing them for mobile devices, the system achieves good detection accuracy with reduced energy consumption
Solution Approach 2:
The system adjusts key parameters of the deep learning model including reducing the number of layers, optimizing batch sizes, and tuning learning rates to be suitable for mobile hardware constraints. These parameter changes enable the algorithm to run efficiently on battery-powered devices while maintaining detection effectiveness
3Measurement precision
If behavioral data is collected for training models, then detection accuracy is improved, but data security risks increase
Solution Approach 1:
The patent segments the training process into local device-level training and centralized model updates. Each device trains locally on its own data, keeping sensitive information local, while only sharing anonymized model updates. This segmentation reduces data security risks while still improving detection accuracy through collective learning
Solution Approach 2:
An intermediary mechanism is introduced that processes and anonymizes training data before any potential sharing or cloud processing. This intermediary layer ensures that even if data is transmitted, the security risks are minimized because the data has been transformed into an anonymized format that cannot be traced back to individuals
Data Source
AI summary
Devices and methods for smartphone impostor detection using behavioral and environmental data are provided. Impostors are attackers who take control of an electronic device (e.g., a smartphone) and gain access to confidential and private information of a legitimate user. Embodiments described herein propose a defense-in-depth mechanism to detect impostors quickly with simple deep learning algorithms, which can achieve better detection accuracy than previous works. Embodiments then consider protecting the privacy of the behavioral and/or environmental data (e.g., collected by one or more sensors) of a user by not exposing it outside the protected device. A recurrent neural network (RNN)-based deep learning algorithm is proposed which uses only sensor data of the legitimate user to learn their normal behavior. Prediction error distribution (PED) is used to enhance the detection accuracy. In some embodiments, a minimalist hardware module, dubbed smartphone impostor detector (SID), is integrated into smartphones for self-contained impostor detection.