notesum.ai

Published at November 22

RED: Effective Trajectory Representation Learning with Comprehensive Information

cs.AI

Released Date: November 22, 2024

Authors: Silin Zhou¹, Shuo Shang¹, Lisi Chen¹, Christian S. Jensen², Panos Kalnis³

Aff.: ¹University of Electronic Science and Technology of China; ²Aalborg University; ³KAUST

Arxiv: http://arxiv.org/abs/2411.15096v1

Refer to caption

	Porto			Rome			Chengdu
	MAE $\downarrow$	MAPE(%) $\downarrow$	RMSE $\downarrow$	MAE $\downarrow$	MAPE(%) $\downarrow$	RMSE $\downarrow$	MAE $\downarrow$	MAPE(%) $\downarrow$	RMSE $\downarrow$
Traj2vec	1.624	20.005	3.137	5.933	45.292	7.188	1.397	16.833	2.114
T2vec	1.608	19.906	3.043	5.873	44.975	7.104	1.400	17.072	2.136
Trembr	1.576	19.391	2.892	5.336	44.220	6.907	1.374	16.635	1.937
PIM	1.684	20.137	3.205	5.992	45.937	7.294	1.622	20.897	2.796
Transformer	1.844	21.325	3.790	6.684	47.131	9.982	1.707	22.458	3.031
BERT	1.707	20.243	3.275	6.114	46.142	7.358	1.453	17.520	2.544
PIM-TF	2.018	22.456	3.898	6.941	48.020	8.313	1.833	23.561	3.163
Toast	1.801	20.952	3.334	6.411	46.864	7.997	1.391	16.984	2.110
START	1.506	18.012	2.799	4.342	43.823	6.282	1.278	15.961	1.889
RED	1.468	16.716	2.651	3.798	41.658	5.358	1.212	15.648	1.720
Improve.	2.52%	7.20%	5.29%	12.53%	4.94%	14.71%	5.16%	1.96%	8.95%