DSHformer: Locality-Sensitive Hash Attention and Prototype Alignment for Sensor-Based Human Activity Recognition
Xiaofeng Zhang, Muzi Ding, Tangzhi Teng, Jie Wan, Hong Ding · 2026
Read DSHformer: Locality-Sensitive Hash Attention and Prototype Alignment for Sensor-Based Human Activity Recognition by Xiaofeng Zhang, Muzi Ding, Tangzhi Teng, Jie Wan, Hong Ding (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13307508.txt.
Description
Xiaofeng Zhang and co-authors present in 2026 DSHformer, a lean model that reads wearable sensor signals to name human actions for health, sport, and home care. Two faults drive the design: shifts in user or sensor place hurt broad use, and full attention grows too costly on long signal streams.
The model packs channel and time traits into small patches, aligns traits to learned class prototypes, and uses hashing based attention that keeps like signals grouped. Scores on five public data sets ran from 88.5 percent up to 98.6 percent across the sets tested here. Tests that removed parts one by one showed each part added value to the whole.
Catalog record
| CREATOR | Xiaofeng Zhang, Muzi Ding, Tangzhi Teng, Jie Wan, Hong Ding |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69615 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |