论文部分内容阅读
深度学习作为机器学习领域的新课题,在学术界和工业界引起了广泛关注,掀起了大数据与人工智能发展的新浪潮。深度学习通过模拟人脑的分层结构,建立了从底层到高层逐级提取输入数据特征的模型,能够深刻揭示从底层信号到高层语义的映射关系。文章从深度学习在互联网、语音图像处理等领域取得的显著成就出发,介绍了深度学习的理论框架,详细阐述了深度学习最为关键的训练过程,概述了三种典型的深度学习模型,包括自动编码器模型、受限玻尔兹曼机模型和深信度网络模型,最后探讨了深度学习所面临的机遇和挑战,以及有待进一步研究解决的问题。
Deep learning, as a new topic in the field of machine learning, has drawn wide attention in academia and industry, setting off a new wave of big data and artificial intelligence development. Depth Learning By modeling the human brain’s hierarchical structure, a model of step-by-step extraction of input data characteristics from bottom to top is established, which can deeply reveal the mapping from the bottom signal to the top semantics. Based on the deep achievements made in the fields of Internet and speech image processing, the article introduces the theoretical framework of deep learning, elaborates the most crucial training process of deep learning in detail, and summarizes three typical deep learning models including automatic coding Model, the restricted Boltzmann model and the model of the deep belief network. Finally, the opportunities and challenges that deep learning faces and the issues to be further studied are discussed.