notesum.ai
Published at October 18ST-MoE-BERT: A Spatial-Temporal Mixture-of-Experts Framework for Long-Term Cross-City Mobility Prediction
cs.CL
cs.AI
cs.CY
Released Date: October 18, 2024
Authors: Haoyu He1, Haozheng Luo2, Qi R. Wang1
Aff.: 1Northeastern University, Boston, MA, USA; 2Northwestern University, Evanston, IL, USA

| Method | A | B | C | D | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GEO-BLEU | DTW | Acc. | GEO-BLEU | DTW | Acc. | GEO-BLEU | DTW | Acc. | GEO-BLEU | DTW | Acc. | |
| HF | 0.266 | 80.3 | 20.4% | 0.265 | 56.4 | 21.0% | 0.251 | 42.4 | 20.8% | 0.295 | 80.0 | 21.0% |
| BERT | 0.256 | 35.7 | 23.8% | 0.284 | 20.6 | 27.0% | 0.253 | 65.6 | 18.2% | 0.253 | 65.6 | 18.2% |
| ST-MoE-BERT w/o PT | 0.286 | 30.2 | 27.9% | 0.286 | 28.2 | 27.5% | 0.294 | 20.7 | 27.9% | 0.250 | 67.6 | 21.4% |
| ST-MoE-BERT | - | - | - | 0.297 | 29.3 | 28.7% | 0.297 | 19.7 | 28.9% | 0.300 | 48.1 | 26.5% |