深度学习理论——第一周homework

Week 1-introduction to deep learning

1.what does the analogy"AI is the new electricity"refer to?
(和“AI是新电力”相类似的说话是什么？)

• Ai is powering personal devices in our homes and offices,similar to electricity.(AI为我们的家庭和办公室个人设备共带你，类似于电能)
• Through the “smart grid”，Ai is delivering a new wave of electricity（通过"智能电网"，AI提供新的电能）
• Ai runs on computers and is thus powered by electricity,but it is letting computers do things not possible before(Ai在计算机上运行，并有电力驱动，但是它正在让以前的计算机不能做的事情变为可能)
• similar to electricity starting about 100 years ago，Ai is transforming multiple industries.(就像100年前产生电能一样，AI正在改变很多行业。)

2.Which of these are reasons for Deep Learning recently taking off?(Check the two options that apply)（哪些是深度学习快速发展的原因？）（两个选项）

• we have access to a lot more computational power(我们有了更好更快的计算能力)
• Neural Networks are a brand new field.(神经网络是一个全新的领域)
• We have access to a lot more data(我们现在可以获取更多的数据)
• Deep learing has resulted in significant improvements in important applications such as online adversting,speech recognition,and image recognition.(深度学习已经取得了重大的进展，比如在在线广告、语音识别和图像识别方面有了很多的应用。)

3.Recall this diagram of iterating over different ML ideas.Which of the statements below are true?(回想一下关于不同的机器学习思想的迭代图。下面那个陈述是正确的？)

• Being able to try out ideas quicklyu allows deep learning engineers to iterate more quickly.(能够让深度学习工程快速地实现自己的想法。)
• Faster computation can help speed up how long a team takes to iterate to a good idea(在更快更好的计算机上能够帮助一个团队减少迭代（训练）的时间)
• It is faster to train on a big dataset than small dataset.(在数据量很多的数据集上训练上的时间要快于小数据集，)
• Recent progress in deep learning algorithms has allowed us to train on a same model.（使用更新的深度学习算法可以使我们能够更快地训练好模型）
  Note： A bigger dataset generally requires more time to train on a same model. (请注意：同一模型在较大数据集上通常需要花费更多时间)

4. WHen an experinced deep learning engineer works on a new problem,they can usually use insight from previous problems to train a good model on the first try,without needing to iterate mulitple times through different models.True/False?
   (当一个经验丰富的深度学习工程师在处理一个新的问题的时候，他们通常可以利用先前的经验来在第一次尝试中训练一个表现很好的模型，而不需要通过不同的模型迭代多次从而选择一个较好的模型，这个说法是正确的还是错误的？)

• True
• False

5. Which one of the plots represents a ReLU activation function？（这些图中的哪一个表示RELU激活功能？）
   

6.Images for cat recognition is an example of of “Structured” data,because it is represented as a structured in a computer.True/False?(用于识别猫的图像是“结构化”数据的一个例子，因为他在计算机中被表示为结构化矩阵，是真的吗？)

• True
• False

图像声音序列其实都不是结构化数据，相比房屋价格预测这种结构化的数据，计算机处理起来相对困难

7.A demographic dataset with statistics on different cities population,GDP per capita,economic growth is an example of "unstructured" data becauser it contains data coming from different
sources.True/False?(统计不同城市人口、人均GDP、经济增长的人口统计数据集是“非结构化”数据的一个例子，因为他包含来自不同来源的数据，是真的吗？)

• True
• False

8.Why is an RNN used for machine translation,say translating English to French?(为什么在上RNN可以应用机器翻译将英语翻译成法语？)

• It can be trained as supervised learining problem.(因为它可以被用作监督学习)
• It is strictly more powerful than a Convolutional Neural Network(CNN).(严格意义上它比卷积神经网络效果要好)
• It is applicable when the input/output is a sequence.(他比较适合用于一个当输入/输出是一个序列的时候)
• RNNs represent the recurrent process of idea->Code -> Experiment ->......(RNNS 代表递归过程：想法->编码->实验->想法->.....)

ANSWER（回答）：
x-axis is the amount of data(x轴是数据量)
y-axis(vertical axis)is the performance of the algorithm(y轴是算法表现的性能)

10. Assuming the trends described in the previous question's figure
    are accurate(and hoping you got the axis labels right),which of the following are true?(check all that apply)（假设上一个问题图中描述的是准确的（并且希望您的轴标签正确），下面哪项是正确的？）

• Increasing the training set generally does not hurt an algorithm performance,and it may help significantly.(增加训练集的大小通常不会影响算法的性能，这可能会有很大帮助。)

• Increasing the size of a neural network generally does not hurt an algorithm performance,and it may help significantly.(增加神经网络的大小通常不会影响算法的性能，这可能会有很大的帮助)

• Decreasing the training set size generally does not hurt an algorithm performance, and it may help significantly.(减少训练集的大小通常不会影响算法的性能，这可能会有很大的帮助。)

• Decreasing the size of a neural network generally does not hurt an algorithm performance,and it may help significantly.(减少神经网络的大小通常不会影响算法的性能，这可能会有很大的帮助)

总结：

结构化数据和非结构化数据，我们都知道传统计算机算法处理的可能都是结构化数据效果比较好，而处理起来非结构化数据相对困难。那么什么是结构化数据呢？吴恩达描述为数据的数据库，其实也就是对于每个特征的描述都有十分清晰定义，比如面积的位置预测房价。然后对于图像处理，特征并没有清晰的定义，需要深度学习算法自己学习出特征信息，最终得到一个函数可以来识别图像特征，也就是训练得到的神经网络