Rights statement: Copyright c 2018, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.
Accepted author manuscript, 1.04 MB, PDF document
Available under license: CC BY-NC: Creative Commons Attribution-NonCommercial 4.0 International License
Final published version
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
Research output: Contribution in Book/Report/Proceedings - With ISBN/ISSN › Conference contribution/Paper › peer-review
}
TY - GEN
T1 - Temporal-difference Learning with Sampling Baseline for Image Captioning
AU - Chen, Hui
AU - Ding, Guiguang
AU - Zhao, Sicheng
AU - Han, Jungong
N1 - Copyright c 2018, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.
PY - 2018/2/1
Y1 - 2018/2/1
N2 - The existing methods for image captioning usually train the language model under the cross entropy loss, which results in the exposure bias and inconsistency of evaluation metric. Recent research has shown these two issues can be well addressed by policy gradient method in reinforcement learning domain attributable to its unique capability of directly optimizing the discrete and non-differentiable evaluation metric. In this paper, we utilize reinforcement learning method to train the image captioning model. Specifically, we train our image captioning model to maximize the overall reward of the sentences by adopting the temporal-difference (TD) learning method, which takes the correlation between temporally successive actions into account. In this way, we assign different values to different words in one sampled sentence by a discounted coefficient when back-propagating the gradient with the REINFORCE algorithm, enabling the correlation between actions to be learned. Besides, instead of estimating a "baseline" to normalize the rewards with another network, we utilize the reward of another Monte-Carlo sample as the "baseline" to avoid high variance. We show that our proposed method can improve the quality of generated captions and outperforms the state-of-the-art methods on the benchmark dataset MS COCO in terms of seven evaluation metrics.
AB - The existing methods for image captioning usually train the language model under the cross entropy loss, which results in the exposure bias and inconsistency of evaluation metric. Recent research has shown these two issues can be well addressed by policy gradient method in reinforcement learning domain attributable to its unique capability of directly optimizing the discrete and non-differentiable evaluation metric. In this paper, we utilize reinforcement learning method to train the image captioning model. Specifically, we train our image captioning model to maximize the overall reward of the sentences by adopting the temporal-difference (TD) learning method, which takes the correlation between temporally successive actions into account. In this way, we assign different values to different words in one sampled sentence by a discounted coefficient when back-propagating the gradient with the REINFORCE algorithm, enabling the correlation between actions to be learned. Besides, instead of estimating a "baseline" to normalize the rewards with another network, we utilize the reward of another Monte-Carlo sample as the "baseline" to avoid high variance. We show that our proposed method can improve the quality of generated captions and outperforms the state-of-the-art methods on the benchmark dataset MS COCO in terms of seven evaluation metrics.
KW - Image captioning
KW - Reinforcement learning
KW - LSTM
M3 - Conference contribution/Paper
SN - 9781577358008
SP - 6706
EP - 6713
BT - 32nd AAAI Conference on Artificial Intelligence 2018
PB - AAAI
CY - Palo Alto
ER -