Baiyu (Breeze) Chen

CS PhD Candidate @UNSW | Building Egocentric Multimodal Intelligence

[Baiyu (Breeze) Chen @Br33z3lled]

6/6 Resources 🔗
Paper: https://t.co/qCSxFLEn7P
Code: https://t.co/Z4rbq4B8VA
Model: https://t.co/rLeEsZM03N
Benchmark: https://t.co/ZRDYaDwHpK
See you in Sydney this December! 🇦🇺

[Baiyu (Breeze) Chen @Br33z3lled]

Excited to share our work AnyMo: Geometry-Aware Setup-Agnostic Modeling of Human Motion in the Wild, accepted at #NeurIPS2026! 🎉

AnyMo enables IMU understanding across different sensor setups and datasets for recognition, retrieval, and captioning.
https://t.co/T3xYNP5lez
🧵⬇️

[Baiyu (Breeze) Chen @Br33z3lled]

5/6 We also release AnyMo-Bench: 211.6 hours of real-world IMU data from 196 participants, covering 150 fine-grained daily activities.
To our knowledge, it is the most fine-grained real-world IMU HAR dataset and benchmark, supporting unseen-subject and cross-device evaluation.

[Baiyu (Breeze) Chen @Br33z3lled]

4/6 Across 14 unseen HAR datasets, AnyMo improves average Accuracy/F1/R@2 by 11.7%/11.6%/22.6%.
It also improves IMU→Text/Text→IMU retrieval MRR by 15.9%/28.6%, and zero-shot motion-captioning BERT-F1 by 18.8%.

[Baiyu (Breeze) Chen @Br33z3lled]

3/6 AnyMo compresses multi-position IMU streams into compact full-body motion tokens and aligns them with an LLM, enabling:
• zero-shot HAR
• IMU↔text retrieval
• wearable motion captioning

[Baiyu (Breeze) Chen @Br33z3lled]

2/6 AnyMo simulates plausible IMU signals at 2,374 body-surface placements, then uses masked cross-view predictive-contrastive learning to recover full-body motion from sparse, setup-specific observations.

[Baiyu (Breeze) Chen @Br33z3lled]

1/6 Wearable IMU signals vary substantially with body location, mounting orientation, hardware, and sampling protocol.
Our key insight: setup variation is not arbitrary noise, it is structured by body-surface geometry, motion, and sensor physics.

[Baiyu (Breeze) Chen @Br33z3lled]

The workshop will include paper presentations, demos/discussions and a Best Paper Award.

Last year in Espoo, UbiSense had 75+ attendees. This year, IMWUT editor in chief Prof. Thomas Ploetz will join as a keynote speaker.

CFP: https://t.co/cipVpuPzjp

[Baiyu (Breeze) Chen @Br33z3lled]

📢 Deadline extended:UbiComp 2026 UbiSense Workshop

We welcome broad UbiComp × emerging tech work: LLMs, agents, foundation models, wearable/mobile sensing, multimodal sensing, edge AI, privacy, datasets & benchmarks.

🗓 Jun 27, 2026 AoE
📍 Shanghai @ UbiComp/ISWC 2026
#UbiComp

[Baiyu (Breeze) Chen @Br33z3lled]

Long papers: up to 6 pages
Short papers: up to 4 pages

Accepted papers will appear in the UbiComp/ISWC 2026 Adjunct Proceedings with DOI.

[Baiyu (Breeze) Chen @Br33z3lled]

Topics include wearables/mobile sensing, smart homes/cities/health, LLMs/agents/foundation models for sensor data, multimodal sensing, vision-language-sensor agents, human-AI/agent collaboration, privacy/trustworthy sensing, edge AI, datasets, benchmarks & evaluation.

[Baiyu (Breeze) Chen @Br33z3lled]

This year UbiSense workshop focuses on emerging techniques in ubiquitous sensing & interaction, especially how AI agents, LLMs and foundation models can be integrated into real-world UbiComp sensing and interaction scenarios.