Related Experiment Video
Updated: Dec 29, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.1K
Zero-Shot Human Activity Recognition Using Non-Visual Sensors.
Fadi Al Machot1, Mohammed R Elkobaisi2, Kyandoghere Kyamakya3
1Research Center Borstel-Leibniz Lung Center, 23845 Borstel, Germany.
Sensors (Basel, Switzerland)
|February 9, 2020
Summary
This study introduces a novel zero-shot learning approach for activity recognition using sensor data. It enables the detection of unseen activities by transferring knowledge from known activities via semantic similarity.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Activity recognition using sensor data has advanced significantly.
- Current machine learning models excel at recognizing known activities but struggle with novel ones.
- Recognizing all possible activities in advance is impractical and costly.
Purpose of the Study:
- To develop a method for extending learning models to detect unseen activities without prior specific knowledge.
- To leverage sensor data for discovering new activities not present in the training set.
- To enable zero-shot learning in activity recognition.
Main Methods:
- Utilizing sensor data to discover unseen activities.
- Implementing a zero-shot learning approach.
- Transferring knowledge from seen to unseen activities using semantic similarity.
Main Results:
- The proposed approach shows promising results for zero-shot learning.
- High performance was achieved in recognizing unseen activities.
- Evaluation on CASAS datasets validates the effectiveness of the method.
Conclusions:
- Sensor data can effectively support zero-shot learning for activity recognition.
- The semantic similarity approach enables knowledge transfer for detecting novel activities.
- This method addresses the challenge of recognizing previously unknown activities in real-world settings.
