The third stage unsupervised learning sequence model
[Core] knowledge
- K-means, GMM and EM
- Hierarchical clustering, DCSCAN, Spectral Clustering Algorithm
- Hidden variables and hidden variables model, Partition function
- conditional independence, D-Separation, Markov properties
- HMM based on the Viterbi Decoding
- Forward / Backward algorithm
- EM algorithm to estimate parameters
- differentiated directed graph and undirected graph model
- Log-Linear Model, logistic regression, eigenfunction
- MEMM problem with Label Bias
- Linear CRF and parameter estimation
The fourth stage of deep learning
[Core] knowledge
- neural network with the activation function
- BP algorithm, convolution layer, Pooling layer, fully connected layer
- convolution neural network, CNN common structure
- Dropout与Batch Normalization
- SGD, Adam, Adagrad algorithm
- RNN gradient disappears, LSTM and GRU
- Seq2Seq model and attentional mechanisms
- Word2Vec, Elmo, Bert, XLNet
- the assistant technical depth learning
- Learning and FIG embedding depth (Graph Embedding)
- Translating Embedding (TransE)
- Node2Vec
- Graph Convolutional Network
- Graph Neural Network
- Dynamic Graph Embedding
[Section] Case explain
- Based Machine Translation and attention mechanisms Seq2Seq
- Based on TransE map and knowledge reasoning GCN
- Face Detection Based on the key points of CNN