Bullet Points
Bullet Points
This post lists out various topics under the machine learning subject.
Data
- data modalities
- numbers
- texts
- images
- videos
- audios
- data cleaning
- imbalanced data
- data normalization (one pitfall)
- standardization
- data augmentation
- data splitting
- cross validation
- RANSAC
- feature extraction/engineering
- domain expertise
- mathematical methods
- PCA
- kernel methods
- neural network self-adaptation
Model
- supervised vs. unsupervised
- supervised learning
- fits towards given target
- includes
- classification
- logistic regression
- support vector machine
- Bayes classifier
- linear discriminant analysis
- extrapolation
- from classification model to regression model
- from binary classification to multi-class classification
- regression (linear and non-linear case)
- classification
- unsupervised learning
- discovers inherent structures within/among the data
- includes
- clustering
- anomaly detection
- dimensionality reduction
- manifold embedding
- latent variable modelling
- hidden Markov model
- conditional random field
- independent component analysis
- reinforcement learning
- adapts to the environment via explore-and-exploit strategy
- depending on the knowledge of the environment, categorizes into
- model-based
- model-free
- depending on the learning objective, categorizes into
- value-based (prediction problem)
- policy-based (control problem)
- supervised learning
- discriminative vs. generative
- linear vs. non-linear (depending on whether the decision boundary is a hyperplane)
- parametric vs. non-parametric (depending on whether the data is characterized 1) by a distribution 2) with a finite number of parameters)
- ensemble method
- bootstrap aggregating
- random forest
- gradient boosting
- least square boosting
- AdaBoost
- LogitBoost
- bootstrap aggregating
- regularization and overfitting
Evaluation
training criterion
- mean squared error
- cross entropy
- impurity
- Gini index
- entropy
testing criterion
confusion matrix
Real Positive Negative Positive TP FN Negative FP TN - accuracy
- precision
- recall
- F-score
precision-recall curve