ZARA: Training-Free Motion Time-Series Reasoning via Evidence-Grounded LLM Agents
Zechen Li, Baiyu Chen, Flora D. Salim
Summary
Per-dataset Macro-F1 (ZARA Gemini, Table 1): Opportunity 92.5 · UCI-HAR 90.0 · Shoaib 97.1 · PAMAP2 76.9 · USC-HAD 60.1 · MHealth 86.1 · WISDM 64.1 · DSADS 84.4.
Key facts
- Form Factor
- other
- License
- Not stated
Reported benchmarks
Human activity recognition (IMU)
mhealth · Macro-F1 ↑
same ZARA protocol, 12 classes.
Human activity recognition (IMU)
opportunity · Macro-F1 ↑
ZARA protocol: knowledge base from seen subjects, class-balanced held-out-subject test, 4 classes (Table 1).
Human activity recognition (IMU)
PAMAP2 · Macro-F1 ↑
Held-out-subject class-balanced split, 12 classes — different protocol from the LOSO MorphCL/HARNet row; not ranked against it (protocol isolation). MorphCL LOSO row unchanged
Updated: Oct 12, 2026
Related Research
- Human activity recognition (IMU) leaderboard
Benchmark results for human activity recognition (imu)
- AutoResearcHAR: A Multi-Agent Framework for Dataset Adaptive IMU Based Human Activity Recognition
Treats IMU-HAR as a whole-pipeline optimisation problem run by LLM agents: a Data Agent (source sele
- Lapras: Latent Reasoning for Time Series Language Models
Post-training framework for time-series language models (TSLMs): instead of verbalising the signal i
- REFIT: Recognize, Fix, and Test Wearable Sensor Placement Shifts without Labels
A gradient-free, label-free input calibration that sits in front of a frozen IMU activity classifier