基于华为AI平台的深度学习实践

近年来，深度学习在人工智能的探索过程中大放异彩。它学习样本数据的内在规律和表示层次，最终让机器能够像人一样具有分析学习能力。作为一个复杂的机器学习算法，深度学习在图像处理、机器翻译、语音等领域都具有显著的优势，在AI各大方向的研究中取得了优越的成果。

本课程是学生创新中心联合华为公司打造的一门AI深度学习校企选修课。本课程将基于华为先进的AI平台——ModelArts，带领学生学习Machine Learning算法，掌握Deep Learning模型架构，理解经典卷积神经网络模型CNN与循环神经网络RNN。结合实际语音或图像识别等案例，利用当前主流的Tensorflow/Keras框架编写AI程序，结合华为ModelArts云平台进行模型训练与部署。

本课程多采用理论与实践相结合的学习模式，带领同学们学习主流AI算法的同时也提供了接触企业先进平台的机会。并且，在学习过程中，会为同学们提供免费的在线计算资源，让学生动手去训练模型并把模型应用到实际项目中。

Inrecent years, Deep Learning has made a big splash in the exploration ofArtificial Intelligence. It learns the intrinsic regulation of big data, andultimately allows machines to have analytical learning capabilities likehumans.

As acomplex machine learning algorithm, Deep Learning has significant advantages inthe fields of image processing, machine translation, and so on. It has achievedsuperior results in the research of AI.

Thiscourse is an AI Deep Learning school-enterprise course created by the StudentInnovation Center and Huawei, which is based on Huawei's advanced AI platform——ModelArts. The course will lead students to learn the Machine Learning algorithm, masterthe Deep Learning model, and understand the classical Convolutional Neural Network(CNN) and Recurrent Neural Network(RNN). Combinedwith experiments on voice and image recognition, we use the Tensorflow/Kerasframework to write AI programs. Huawei ModelArts cloud platform is used formodel training and deployment.

Thiscourse combines the theory and practice, to learn the mainstream AI algorithmand also to provide opportunities to access the advanced platform of theenterprise. Moreover, during the course, we will provide free online computingresources for students, allowing them to train and apply the model tothe actual project.

1、理解世界AI态势与中国“科技强国”战略，培养上海交大学生的家国情怀与责任担当。树立正确的工程伦理价值观与职业道德：尊重数据版权，合理合法运用AI算法。

2、系统地掌握回归分类模型、神经网络与BP反向传播原理，熟练运用华为云平台完成分类任务，具备分析、优化网络结构的能力。

3、能阐明CNN卷积神经网络、Inception等进阶网络结构，熟练应用TensorFlow完成图像识别任务。

4、能比较不同目标检测算法的优劣势，能基于华为AI平台进行标注、训练、部署等AI开发任务，掌握综合实验方法。同时理解基础语言模型，了解进阶的BERT模型与Transformer。