Summary of Unsupervised Domain Adaptation For Action Recognition Via Self-ensembling and Conditional Embedding Alignment, by Indrajeet Ghosh et al.
Unsupervised Domain Adaptation for Action Recognition via Self-Ensembling and Conditional Embedding Alignment
by Indrajeet Ghosh, Garvit Chugh, Abu Zaher Md Faridee, Nirmalya Roy
First submitted to arxiv on: 23 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed architecture addresses the limitations of deep learning-based wearable human action recognition (wHAR) by introducing a novel joint optimization framework that improves model classification generalizability and domain generalizability. The architecture consists of three functions: consistency regularizer, temporal ensemble for robust pseudo-label generation, and conditional distribution alignment to minimize kernel-based class-wise conditional maximum mean discrepancy (kCMMD). These components work together to ensure strong generalization with limited source domain samples and consistent pseudo-label generation in target samples. The proposed approach outperforms six state-of-the-art unsupervised domain adaptation methods by achieving an average macro-F1 score improvement of around 4-12% across four benchmark wHAR datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to improve wearable human action recognition (wHAR) using deep learning. The goal is to make it easier for devices like smartwatches and fitness trackers to recognize different actions, like running or dancing. Right now, these devices are not very good at recognizing complex actions because they don’t have enough training data. The paper introduces a new way of combining different versions of the same action (like slightly different recordings) to improve recognition accuracy. This approach also helps to reduce errors caused by differences between how people perform the same action. The results show that this method works better than other methods in recognizing human actions. |
Keywords
» Artificial intelligence » Alignment » Classification » Deep learning » Domain adaptation » F1 score » Generalization » Optimization » Unsupervised
