qriton/hlm-micro-har-v1-uci-har

HLM-Micro HAR v1 - UCI HAR

HLM-Micro HAR v1 is a compact polynomial-Hopfield classifier trained on the UCI Human Activity Recognition dataset.

Results

Field	Value
Parameters	277,760
Classes	6 activities
Reported test accuracy	93.52%
Test windows	2,947 held-out UCI HAR windows
Data	Smartphone IMU activity windows from 30 subjects

Per-class summary from the source evaluation:

Class	Accuracy
Walking	99.60%
Upstairs	98.09%
Downstairs	93.10%
Sitting	90.84%
Standing	95.30%
Laying	84.92%

Files

File	Purpose
model.pt	Sanitized model-only PyTorch checkpoint
config.json	Public architecture, task, classes, and metric metadata
metrics.jsonl	Training/evaluation metrics from the local run

Intended Use

• Research on compact human-activity-recognition models.
• TinyML and edge-sensor experiments.
• Baseline for HLM-Micro architecture comparisons.

Limitations

• UCI HAR is a fixed benchmark, not a complete wearable deployment test.
• Not a medical, fall-detection, gait-analysis, or safety-critical model.
• Production use would require new data collection, calibration, and deployment testing.