Cross-Body Transfer Learning for Human Activity Recognition
Bibliographic record
Abstract
Human activity recognition (HAR) is the field of using analysis of sensor recordings to detect and determine what activity a person is doing. At the lowest level, it detects simple movements like standing, sitting, walking, and others. These movements are key components for measuring a person’s health and are the subject of many clinical assessments. Devices like hearing aids are the perfect target for HAR from head-worn sensors as many older adults wear and use them regularly. A major constraint for head-worn human-activity-recognition (HAR) deep learning models is gathering enough data. There are many HAR accelerometer datasets recorded from other parts of the body which may fill this gap. In this paper, we show that we can use cross-body transfer learning to improve HAR classification on the UCA-E-HAR head dataset which was recorded using smart glasses. To show this, we gathered datasets from the ankle, thigh, lower back, chest, wrist, and pocket to be used for transfer learning. The datasets included are KU-HAR, Forth, GOTOV, HARTH, Motionsense, Har70+, and UCA-E-HAR. We found that the classification using Resnetv1-6 model for the head data performance improved by 8% to 84.8% for transfer learning models pretrained on all available data from the other body locations. and the models were also better at differentiating between similar tasks such as standing and sitting. We then showed that removing standing from the possible movements, due to its similarity to sitting, allowed the base model performance to increase from 76.8% to 90.8% for a model trained using only head data. The transfer learning model trained on all available data improved performance to 94.5%. This shows that there are benefits for performing cross-body transfer learning for HAR classification.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".