Fundamentals of machine learning

• Supervised learning
  • binary classification
  • multiclass classificaiton
  • scalar regression
  • vector regression（比如bounding-box)
  • Sequence generation (摘要，翻译...)
  • Syntax tree prediction
  • Object detection (一般bounding-box的坐标仍然是回归出来的)
  • Image segmentation
• Unsupervised learing
  • 是数据分析的基础，在监督学习前也常常需要用无监督学习来更好地“理解”数据集
  • 主要有降维(Dimensionality reduction)和聚类(clustering)
• Self-supervised learning
  • 其实还是监督学习，因为它仍需要与某个target做比较
  • 往往半监督（自监督）学习仍然有小量有标签数据集，在此基础上训练的不完善的model用来对无标签的数据进行打标，循环中对无标签数据打标的可靠度就越来越高，这样总体数据集的可靠度也越来越高了。有点像生成对抗网络里生成器和辨别器一同在训练过程中完善。
  • autoencoders
• Reinforcement learning
  • an agent receives information about its environment and learns to choose actions that will maximize some reward.
  • 可以用训练狗来理解
  • 工业界的应用除了游戏就是机器人了

Data preprocessing

• vectorization
• normalization (small, homogenous)
• handling missing values
  1. 除非0有特别的含义，不然一般可以对缺失值补0
  2. 你不能保证测试集没有缺失值，如果训练集没看到过缺失值，那么将不会学到忽略缺失值
     • 复制一些训练数据并且随机drop掉一些特征
• feature extraction
  • making a problem easier by expressing it in a simpler way. It usually requires understanding the problem in depth.
  • Before deep learning, feature engineering used to be critical, because classical shallow algorithms didn’t have hypothesis spaces rich enough to learn useful features by themselves. (又见假设空间)
  • 但是好的特征仍然能让你在处理问题上更优雅、更省资源，也能减小对数据集规模的依赖。

Overfitting and underfitting

• Machine learning is the tension between optimization and generalization.
• optimization要求你在训练过的数据集上能达到最好的效果
• generalization则希望你在没见过的数据上有好的效果
• 如果训练集上loss小，测试集上也小，说明还有优化(optimize)的余地 -> underfitting看loss
  • just keep training
• 如果验证集上generalization stop improving(泛化不再进步，一般看衡量指标，比如准确率) -> overfitting

解决overfitting的思路：

• the best solution is get more trainging data
• the simple way is to reduce the size of the model
  • 模型容量(capacity)足够大，就足够容易记住input和target的映射，没推理什么事了
• add constraints -> weight regularization
• add dropout

Regularization

Occam’s razor

given two explanations for something, the explanation most likely to be correct is the simplest one—the one that makes fewer assumptions.

即为传说中如无必要，勿增实体的奥卡姆剃刀原理，这是在艺术创作领域的翻译，我们这里还是直译的好，即能解释一件事的各种理解中，越简单的，假设条件越少的，往往是最正确的，引申到机器学习，就是如何定义一个simple model

A simple model in this context is:

• a model where the distribution of parameter values has less entropy
• or a model with fewer parameters

实操就是，就是迫使选择那些值比较小的weights，which makes the distribution of weight values more regular. This is called weight regularization。这个解释是我目前看到的最regularization这个名字最好的解释，“正则化”三个字都认识，根本没人知道这三个字是什么意思，翻译了跟没番一样，而使分布更“常规化，正规化”，好像更有解释性。

别的教材里还会告诉你这里是对大的权重的惩罚（设计损失函数加上自身权重后，权重越大，loss也就越大，这就是对大权重的惩罚）

• L1 regularization—The cost added is proportional to the absolute value of the weight coefficients (the L1 norm of the weights).
• L2 regularization—The cost added is proportional to the square of the value of the weight coefficients (the L2 norm of the weights).

L2 regularization is also called weight decayin the context of neural networks. Don’t let the different name confuse you: weight decay is mathematically the same as L2 regularization.

只需要在训练时添加正则化

Dropout

randomly dropping out (setting to zero) a number of output features of the layer during training.

dropout的作者Geoff Hinton解释dropout的灵感来源于银行办事出纳的不停更换和移动的防欺诈机制，可能认为一次欺诈的成功实施需要员工的配合，所以就尽量降低这种配合的可能性。于是他为了防止神经元也能聚在一起”密谋”，尝试随机去掉一些神经元。以及对输出添加噪声，让模型更难记住某些patten。

The universal workflow of machine learning

1. Defining the problem and assembling a dataset
   • What will your input data be?
   • What are you trying to predict?
   • What type of problem are you facing?
   • You hypothesize that your outputs can be predicted given your inputs.
   • You hypothesize that your available data is sufficiently informative to learn the relationship between inputs and outputs.
   • Just because you’ve assembled exam- ples of inputs X and targets Y doesn’t mean X contains enough information to predict Y.
2. Choosing a measure of success
   • accuracy? Precision and recall? Customer-retention rate?
   • balanced-classification problems,
     • accuracy and area under the receiver operating characteristic curve (ROC AUC)
   • class-imbalanced problems
     • precision and recall.
   • ranking problems or multilabel classification
     • mean average precision
   • ...
3. Deciding on an evaluation protocol
   • Maintaining a hold-out validation set—The way to go when you have plenty of data
   • Doing K-fold cross-validation—The right choice when you have too few samples for hold-out validation to be reliable
   • Doing iterated K-fold validation—For performing highly accurate model evaluation when little data is available
4. Preparing your data
   • tensor化，向量化，归一化等
   • may do some feature engineering
5. Developing a model that does better than a baseline
   • baseline:
     • 基本上是用纯随机(比如手写数字识别，随机猜测为10%)，和纯相关性推理（比如用前几天的温度预测今天的温度，因为温度变化是连续的），不用任何机器学习做出baseline
   • model:
     • Last-layer activation
       • sigmoid, relu系列， 等等
     • Loss function
       • 直接的预测值真值的差，如MSE
       • 度量代理，如crossentropy是ROC AUC的proxy metric
   • Optimization configuration
     • What optimizer will you use? What will its learning rate be? In most cases, it’s safe to go with rmsprop and its default learning rate.
   • Scaling up: developing a model that overfits
     • 通过增加layers, 增加capacity，增加training epoch来加速overfitting，从而再通过减模型和加约束等优化
   • Regularizing your model and tuning your hyperparameters
     • Add dropout.
     • Try different architectures: add or remove layers.
     • Add L1 and/or L2 regularization.
     • Try different hyperparameters (such as the number of units per layer or the learning rate of the optimizer) to find the optimal configuration.
     • Optionally, iterate on feature engineering: add new features, or remove features that don’t seem to be informative.