Please enable JavaScript.
Coggle requires JavaScript to display documents.
Machine Learning (Statistical Model (Conditional Random Field (CRF): used…
Machine Learning
Statistical Model
Conditional Random Field (CRF): used for structure prediction, can take context into account, popular in NLP.
-
Hidden Markov Model (HMM): can be presented as simplest dynamic Bayesian network, model is Markov process with unobserved state*.
Bayesian Inference: User Bayes theorem, use prior probability to update posterior probability
-
-
Neural Network
Convolutional Neural Network:User convolution layer to find features that reside in local group of data. Usually applied to image processing, currently the best.
Auto Encoder: Input and output layers has the same number of nodes, hidden layer has fewer nodes to compress data with fewer info
Function: user layers of many computational nodes, model the neuron in human brain, to learn features from data and calculate the output.
It's very powerful because many layers can represents very complex model.
Recurrent Neural Network (RNN): Connections of the nodes in model form a directed cycle. Can be used to model dynamic memory when learning.
Dimensional Reduction
-
-
Function: reduce dimension of data, easier to visualize, extract features best represent the data
Clustering
-
Spectral Clustering: User eigenvalues of similarity matrix to cluster data in lower dimension. #
Function: Automatically group data into clusters, to find structure inside the data