The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Kaedah ramalan Markov tradisional bagi destinasi teksi hanya bergantung pada 2 hingga 3 mata GPS sebelumnya. Mereka mengabaikan pergantungan jangka panjang dalam trajektori teksi. Kami menggunakan Rangkaian Neural Berulang (RNN) untuk meneroka kebergantungan jangka panjang untuk meramalkan destinasi teksi kerana berbilang lapisan tersembunyi RNN boleh menyimpan kebergantungan ini. Walau bagaimanapun, lapisan tersembunyi RNN sangat sensitif kepada gangguan kecil untuk mengurangkan ketepatan ramalan apabila jumlah trajektori teksi semakin meningkat. Untuk meningkatkan ketepatan ramalan destinasi teksi dan mengurangkan masa latihan, kami membenamkan zon keluar dipacu suprisal (SDZ) kepada RNN, justeru kaedah ramalan destinasi teksi oleh RNN biasa dengan SDZ (TDPRS). SDZ bukan sahaja boleh meningkatkan keteguhan TDPRS, tetapi juga mengurangkan masa latihan dengan menggunakan kemas kini separa parameter dan bukannya kemas kini penuh. Eksperimen dengan data trajektori teksi Porto menunjukkan bahawa TDPRS meningkatkan ketepatan ramalan sebanyak 12% berbanding kaedah ramalan RNN dalam literatur[4]. Pada masa yang sama, masa ramalan dikurangkan sebanyak 7%.
Lei ZHANG
China University of Mining and Technology
Guoxing ZHANG
China University of Mining and Technology
Zhizheng LIANG
China University of Mining and Technology
Qingfu FAN
China University of Mining and Technology
Yadong LI
China University of Mining and Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Salinan
Lei ZHANG, Guoxing ZHANG, Zhizheng LIANG, Qingfu FAN, Yadong LI, "Predicting Taxi Destination by Regularized RNN with SDZ" in IEICE TRANSACTIONS on Information,
vol. E101-D, no. 8, pp. 2141-2144, August 2018, doi: 10.1587/transinf.2018EDL8009.
Abstract: The traditional Markov prediction methods of the taxi destination rely only on the previous 2 to 3 GPS points. They negelect long-term dependencies within a taxi trajectory. We adopt a Recurrent Neural Network (RNN) to explore the long-term dependencies to predict the taxi destination as the multiple hidden layers of RNN can store these dependencies. However, the hidden layers of RNN are very sensitive to small perturbations to reduce the prediction accuracy when the amount of taxi trajectories is increasing. In order to improve the prediction accuracy of taxi destination and reduce the training time, we embed suprisal-driven zoneout (SDZ) to RNN, hence a taxi destination prediction method by regularized RNN with SDZ (TDPRS). SDZ can not only improve the robustness of TDPRS, but also reduce the training time by adopting partial update of parameters instead of a full update. Experiments with a Porto taxi trajectory data show that TDPRS improves the prediction accuracy by 12% compared to RNN prediction method in literature[4]. At the same time, the prediction time is reduced by 7%.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2018EDL8009/_p
Salinan
@ARTICLE{e101-d_8_2141,
author={Lei ZHANG, Guoxing ZHANG, Zhizheng LIANG, Qingfu FAN, Yadong LI, },
journal={IEICE TRANSACTIONS on Information},
title={Predicting Taxi Destination by Regularized RNN with SDZ},
year={2018},
volume={E101-D},
number={8},
pages={2141-2144},
abstract={The traditional Markov prediction methods of the taxi destination rely only on the previous 2 to 3 GPS points. They negelect long-term dependencies within a taxi trajectory. We adopt a Recurrent Neural Network (RNN) to explore the long-term dependencies to predict the taxi destination as the multiple hidden layers of RNN can store these dependencies. However, the hidden layers of RNN are very sensitive to small perturbations to reduce the prediction accuracy when the amount of taxi trajectories is increasing. In order to improve the prediction accuracy of taxi destination and reduce the training time, we embed suprisal-driven zoneout (SDZ) to RNN, hence a taxi destination prediction method by regularized RNN with SDZ (TDPRS). SDZ can not only improve the robustness of TDPRS, but also reduce the training time by adopting partial update of parameters instead of a full update. Experiments with a Porto taxi trajectory data show that TDPRS improves the prediction accuracy by 12% compared to RNN prediction method in literature[4]. At the same time, the prediction time is reduced by 7%.},
keywords={},
doi={10.1587/transinf.2018EDL8009},
ISSN={1745-1361},
month={August},}
Salinan
TY - JOUR
TI - Predicting Taxi Destination by Regularized RNN with SDZ
T2 - IEICE TRANSACTIONS on Information
SP - 2141
EP - 2144
AU - Lei ZHANG
AU - Guoxing ZHANG
AU - Zhizheng LIANG
AU - Qingfu FAN
AU - Yadong LI
PY - 2018
DO - 10.1587/transinf.2018EDL8009
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E101-D
IS - 8
JA - IEICE TRANSACTIONS on Information
Y1 - August 2018
AB - The traditional Markov prediction methods of the taxi destination rely only on the previous 2 to 3 GPS points. They negelect long-term dependencies within a taxi trajectory. We adopt a Recurrent Neural Network (RNN) to explore the long-term dependencies to predict the taxi destination as the multiple hidden layers of RNN can store these dependencies. However, the hidden layers of RNN are very sensitive to small perturbations to reduce the prediction accuracy when the amount of taxi trajectories is increasing. In order to improve the prediction accuracy of taxi destination and reduce the training time, we embed suprisal-driven zoneout (SDZ) to RNN, hence a taxi destination prediction method by regularized RNN with SDZ (TDPRS). SDZ can not only improve the robustness of TDPRS, but also reduce the training time by adopting partial update of parameters instead of a full update. Experiments with a Porto taxi trajectory data show that TDPRS improves the prediction accuracy by 12% compared to RNN prediction method in literature[4]. At the same time, the prediction time is reduced by 7%.
ER -