关于这门课
这门课的授课老师是号称谢菲尔德大学最受欢迎的Lecture - Dr Sean Anderson。课堂的风格确实比较生动有趣,而且因为Deep Learning目前仍然比较前沿的原因,课堂上时常举一些比较cool或者exciting的例子,比如一次关于CUDA的课堂,开始跟我们讨论比特币和RTX3090的价格...
他的口头禅是:Are you happy with that?其实我们大部分时候,对Sean都是比较happy的,但是他说出这句话的语境,往往是大家对他所讲内容一脸懵逼的时候......
这门课共16个Lecture:
Lecture 1 Introduction to Deep Learning
Lecture 2 Machine Learning Basics
Lecture 3 Deep Networks
Lecture 4 Convolutional Networks
Lecture 5 Optimisation
Lecture 6 Regularisation
Lecture 7 Software and Hardware
Lecture 8 Practical methods
Lecture 9 Intro to Reinforcement Learning
Lecture 10 Tabular Reinforcement Learning
Lecture 11 Deep Q-learning
Lecture 12 Policy Gradient, Actor-Critic Methods
Lecture 13 Simple Recurrent Networks
Lecture 14 Gated Recurrent + Echo State Nets
Lecture 15 Generative Adversarial Nets
Lecture 16 State of the Art
深度学习
深度学习在数据非结构化和特征不明显的领域彻底改变了机器学习。
深度学习可以应用于
图像处理 image processing
- 图像分类
- 图像分割
- 实时物体检测和识别
- DeepFake
自然语言处理 natural language processing
- 语音识别
- 语音生成
强化学习 Deep reinforcement learning
- 阿尔法Go
- 星际争霸II的AI电脑
一些著名的深度学习数据集:
Images and video
• Imagenet: Object recognition http://www.image-net.org/
• CIFAR10 & CIFAR100 https://www.cs.toronto.edu/~kriz/cifar.html
• Microsoft COCO: Object detection and recognition https://cocodataset.org/#home
• MNIST: Digit recognition http://yann.lecun.com/exdb/mnist/
• Kitti: Robotics computer vision http://www.cvlibs.net/datasets/kitti/
Speech and audio
• Audioset: Google sound recognition https://research.google.com/audioset/
• Speech command data set: Google speech commands
http://download.tensorflow.org/data/speech_commands_v0.01.tar.gz
Text
• WMT14: Machine (language) translation http://statmt.org/wmt14/translation-task.html
• IMDB: Sentiment analysis http://ai.stanford.edu/~amaas/data/sentiment