Keystroke Learning Algorithm Security for Mobile Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Input string learning algorithms in communication devices can compromise security by storing sensitive information in less secure memory areas, allowing unauthorized users to reconstruct secure messages if the device is lost or stolen.
Innovation Solution
Implementing a user-selectable option to temporarily disable the input string learning algorithm when sensitive information is being input, with automatic reactivation upon completion, and providing notifications for the status of the learning algorithm's activation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the input string learning algorithm is enabled for all user inputs, then the convenience and time-saving predictive text functionality is improved, but the security of sensitive information is compromised
Solution Approach 1:
The learning algorithm is made dynamic by allowing it to be enabled or disabled based on the security context of the input field. The system automatically adjusts its behavior - learning when appropriate (e.g., casual messaging) and not learning when security is concerned (e.g., passwords, sensitive data), thus adapting to different operational conditions.
Solution Approach 2:
A security determination mechanism acts as an intermediary between the input field and the learning algorithm. This intermediary evaluates whether the current input context is secure enough for learning, and only allows the learning algorithm to process inputs that pass the security assessment, thereby mediating between convenience and security requirements.
2Measurement precision
If the learning algorithm stores all entered strings in the dictionary, then the predictive completion accuracy is improved, but the vulnerability to unauthorized reconstruction of secure messages is worsened
Solution Approach 1:
Different quality standards are applied to different portions of the input data based on their security classification. Sensitive information (passwords, personal data) is excluded from the learnable dictionary with high security quality, while non-sensitive information can be stored with lower security quality requirements, allowing predictive functionality to operate on safe data while protecting sensitive data.
Solution Approach 2:
The harmful element (sensitive information) is extracted and removed from the set of data that the learning algorithm can store. By identifying and excluding sensitive inputs from the learning process, the system maintains predictive accuracy on non-sensitive data while preventing unauthorized reconstruction of secure information.
3Productivity
If the learning algorithm is always active, then the user experience and input efficiency are improved, but the device's security against loss or theft is worsened
Solution Approach 1:
The learning algorithm transitions from a static always-on state to a dynamic state that responds to security context. The system maintains high input efficiency by keeping the algorithm active for non-sensitive inputs while dynamically suspending learning for sensitive inputs, thus preserving productivity without compromising reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods and apparatus for disabling a learning algorithm resident on a communications device, such as, for example, a mobile wireless communications device are disclosed. A user may be provided with the option of disabling an algorithm that learns user keystrokes in order to provide predictive completion or disambiguation of input strings when the input string may contain sensitive or secure information. Upon completion of entry of the secure or sensitive data, the learning function is reactivated. Disabling of the learning function may also occur automatically based on an application in which data is being entered.