Collaborative Sensing with Interactive Learning using Dynamic Intelligent Virtual Sensors.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Sensor data is increasingly common but challenging to utilize effectively.
• Existing virtual sensors often handle only homogeneous data from fixed sensor groups.
• The Dynamic Intelligent Virtual Sensors (DIVS) concept addresses dynamic, heterogeneous sensor environments.

Purpose of the Study:

• To refine the DIVS concept by integrating interactive machine learning.
• To enable systems to learn from both user input and real-world sensor data.
• To empirically validate DIVS properties, focusing on data labeling budgets and user strategies.

Main Methods:

• Integration of an interactive machine learning mechanism into the DIVS framework.
• Empirical validation of DIVS properties concerning labeled data budget allocation.
• Analysis of proactive labeling strategies for user input.

Main Results:

• DIVS can achieve good accuracy with limited labeled data in dynamic sensor environments.
• Proactive user labeling strategies significantly enhance system performance.
• The refined DIVS model effectively handles heterogeneous sensor data and user interactions.

Conclusions:

• The refined DIVS concept with interactive machine learning is effective for processing dynamic, heterogeneous sensor data.
• Optimized data labeling strategies are crucial for maximizing performance under budget constraints.
• DIVS provides a robust framework for developing intelligent services from diverse sensor streams.