Dual Inference for Machine Learning
Dual Inference for Machine Learning
复制标题
DOI：
10.24963/ijcai.2017/434
复制
发表时间：
2017-08
期刊：
影响因子：
--
通讯作者：
Yingce Xia;Jiang Bian;Tao Qin;Nenghai Yu;Tie-Yan Liu
中科院分区：
文献类型：
--
作者：
Yingce Xia;Jiang Bian;Tao Qin;Nenghai Yu;Tie-Yan Liu
近年来，机器学习在解决许多领域的阿尔蒂智能（AI）任务方面取得了快速发展，包括翻译，语音，图像等。作为一种特定类型的关系，结构二元性确实存在于许多人工智能任务之间，例如从一种语言翻译到另一种语言与其相反的方向，语音识别与语音合成，图像分类与图像生成等。然而，关于如何将这种宝贵的关系利用到人工智能任务的推理阶段的研究很少。在本文中，我们提出了一个通用的双重推理框架，它可以利用现有的模型从两个双重任务，无需重新训练，进行推理的一个单独的任务。对机器翻译、情感分析和图像处理等三对特定双重任务的实证研究表明，双重推理可以显著提高每个任务的性能。
Recent years have witnessed the rapid development of machine learning in solving artificial intelligence (AI) tasks in many domains, including translation, speech, image, etc. Within these domains, AI tasks are usually not independent. As a specific type of relationship, structural duality does exist between many pairs of AI tasks, such as translation from one language to another vs. its opposite direction, speech recognition vs. speech syntheti-zation, image classification vs. image generation, etc. The importance of such duality has been mag-nified by some recent studies, which revealed that it can boost the learning of two tasks in the dual form. However, there has been little investigation on how to leverage this invaluable relationship into the inference stage of AI tasks. In this paper, we propose a general framework of dual inference which can take advantage of both existing models from two dual tasks, without re-training, to conduct inference for one individual task. Empirical studies on three pairs of specific dual tasks, including machine translation, sentiment analysis, and image processing have illustrated that dual inference can significantly improve the performance of each of individual tasks.