人工智能培训

搜索

深度学习论文:基于深度强化学习的图像标题(Image Captioning based on Deep Reinforcement Learning)

[复制链接]
犀牛 发表于 2018-9-14 09:12:21 | 显示全部楼层 |阅读模式
犀牛 2018-9-14 09:12:21 106 0 显示全部楼层
深度学习论文:基于深度强化学习的图像标题(Image Captioning based on Deep Reinforcement Learning)最近,它已经表明,加强学习的政策梯度方法已被用于训练深度端到端系统的自然语言处理任务。更重要的是,由于理解图像内容的复杂性以及用自然语言描述图像内容的各种方式,图像字幕一直是一个需要处理的具有挑战性的问题。据我们所知,大多数最先进的方法遵循顺序模型的模式,例如递归神经网络(RNN)。然而,在本文中,我们提出了一种新的图像字幕结构,具有深度加强学习,以优化图像字幕任务。我们利用两个称为“策略网络”和“价值网络”的网络来协同生成图像标题。实验在Microsoft COCO数据集上进行,实验结果验证了该方法的有效性。
Recently it has shown that the policy-gradient methods for reinforcementlearning have been utilized to train deep end-to-end systems on naturallanguage processing tasks.What's more, with the complexity of understandingimage content and diverse ways of describing image content in natural language,image captioning has been a challenging problem to deal with.To the best ofour knowledge, most state-of-the-art methods follow a pattern of sequentialmodel, such as recurrent neural networks (RNN).However, in this paper, wepropose a novel architecture for image captioning with deep reinforcementlearning to optimize image captioning tasks.We utilize two networks called"policy network" and "value network" to collaboratively generate the captionsof images.The experiments are conducted on Microsoft COCO dataset, and theexperimental results have verified the effectiveness of the proposed method.深度学习论文:基于深度强化学习的图像标题(Image Captioning based on Deep Reinforcement Learning) mPAZNy5QxQnqB99p.jpg
URL地址:https://arxiv.org/abs/1809.04835     ----pdf下载地址:https://arxiv.org/pdf/1809.04835    ----深度学习论文:基于深度强化学习的图像标题(Image Captioning based on Deep Reinforcement Learning)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

犀牛当前离线
新手上路

查看:106 | 回复:0

快速回复 返回顶部 返回列表