Self-supervised human activity recognition pre-training follows power-law scaling in data hours and model size, and user diversity improves scaling more than per-user data volume.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Scaling laws in wearable human activity recognition
Self-supervised human activity recognition pre-training follows power-law scaling in data hours and model size, and user diversity improves scaling more than per-user data volume.