Reproduce “Personalized Activity Recognition with Deep Triplet Embeddings” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
David M. Burns, Cari M. Whyne
This paper presents a method for personalized human activity recognition using deep triplet embeddings and a novel subject triplet loss, which adapts to individual users by learning compact, discriminative representations. The approach demonstrates improved classification accuracy, effective out-of-distribution detection, and generalization to new activities with minimal reference data requirements across multiple datasets.
Key insights
- Personalization through deep embeddings consistently improves Human Activity Recognition (HAR) performance compared to impersonal models and those relying on hand-engineered features.
- Training deep embeddings with the novel 'subject triplet loss' specifically enhances personalized activity recognition, leading to superior performance.
- The proposed personalized approach is practical, requiring minimal reference data (approximately 16 seconds per activity class) and small embedding sizes (down to 8 features) for effective performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.