METIER: A Deep Multi-Task Learning Based Activity and User Recognition Model Using Wearable Sensors
summary
METIER jointly trains activity and user recognition and claims better robustness to new users and activity shifts by transferring knowledge between tasks. Per-window sensor inputs go through a CNN + BiLSTM pipeline, convolutional layers use tensor-decomposition parameter sharing to softly share weights between tasks, and a mutual attention mapping uses user features to weight activity features and vice versa. Evaluation on three public datasets shows METIER outperforms existing AR and UR baselines.
AI-written summary · about
chapters (6)
- 0:00Research motivation
- 1:11Proposed framework
- 1:43Network overview
- 2:27Technical architecture
- 3:47Mutual attention mechanism
- 4:46Experimental results
related talks
06mUnsupervised Human Activity Representation Learning with Multi-task Deep Clustering
28mDeep Learning-Based Human Activity Recognition for Physical Skills...
10mLearning Network-Based Multi-Modal Mobile User Interface Embeddings
16mMyoSign: enabling end-to-end sign language recognition with wearables
09mAn Activity Recognition System for Taking Medicine Using In-The-Wild Data to Promote Medication ...
10mMIND MELD: Personalized Meta-Learning for Robot-Centric Imitation Learning
34mComputer Vision and NLP for Multi-Task Fashion Attrb. Modeling
06mApproaching the Real-World: Supporting Activity Recognition Training with Virtual IMU Data