Metraipolis › archive › dc_36ecfc895dfd

What Does 4-Bit Microscaling Cost a Wearable Activity Recognizer? — arXiv:2611.04185

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.04185
License (as archived): CC0-1.0
Archived: 2026-11-04T21:51:34.950Z
SHA-256 of the archived text: 642769cfb45b7c0417be7d62c4bb98ea05f76836ce334a639049a12e9ded5ce2
Stable URL of this page: https://metraipolis.com/documento/dc_36ecfc895dfd

How to cite

What Does 4-Bit Microscaling Cost a Wearable Activity Recognizer? — arXiv:2611.04185 — arXiv (en), https://arxiv.org/abs/2611.04185 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_36ecfc895dfd (Metraipolis archive dc_36ecfc895dfd, sha256 642769cfb45b7c0417be7d62c4bb98ea05f76836ce334a639049a12e9ded5ce2)

BibTeX

@misc{dc36ecfc895dfd,
  title = {What Does 4-Bit Microscaling Cost a Wearable Activity Recognizer? — arXiv:2611.04185},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04185},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_36ecfc895dfd; sha256 642769cfb45b7c0417be7d62c4bb98ea05f76836ce334a639049a12e9ded5ce2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04185. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Wearable sports sensors can recognize actions on the device itself, where lower-precision numbers reduce model storage and memory traffic. Two 4-bit block-scaled formats, MXFP4 from the Open Compute Project, introduced in 2023, and NVFP4 from NVIDIA, introduced in 2025, are drawing interest. Published evaluations cover general deep-learning benchmarks and large language models; to our knowledge, nobody has yet measured what they cost a small wearable-sensor model. We train a 159,492-parameter 1D CNN to recognize four basketball actions from wrist-accelerometer data in Hang-Time HAR and test it on players it has never seen, using five-fold cross-validation grouped by player. Every comparison between number formats is made fold by fold. In full precision the model reaches 69.26\% macro-F1. Measured TFLite FP16 and INT8 models lose at most 0.16 percentage points. In simulation, 4-bit weights cost 1.45 points with MXFP4 and 0.72 points with NVFP4. When activations are quantized to 4 bits as well, MXFP4 loses 5.14 points and NVFP4 only 0.75, and NVFP4 comes out ahead in all five held-out-player folds. The ranking survives two alternative simulator settings, with NVFP4 leading by 3.49 to 4.40 points. A $2 \times 3$ ablation with 4-bit weights and activations (two block sizes and three scale rules) points to scale selection as the main source of MXFP4's loss. The reference MXFP4 conversion we simulate rounds the scale down, which can pick scales that are too small, so many blocks clip their largest values. Rounding the scale up recovers 3.18 points with 32-value blocks and 3.63 points with 16-value blocks, and switching to an 8-bit floating-point scale, as NVFP4 does, adds another 0.84 and 1.20 points. Block size matters less. Under this round-down rule, 16-value blocks score 0.44 points below 32-value blocks in all five folds; under the other two rules there is no consistent effect.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.04185 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-04T21:38:49.705Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).