Machine learning high-level theoretical knowledge list

Source: https://mp.weixin.qq.com/s/1MM0_wk43WARz4c5WWuzCQ

Personal learning collection, intrusion and deletion

The purpose of this article is to serve as a reference for the learning route of high-level machine learning theory. Individuals need to make judgments and choices according to the actual situation during personal learning.

Course goal: train more high-end talents and help those who are about to or are currently engaged in scientific research or have been engaged in the AI ​​industry to improve their technical depth. 

The knowledge list is selected from the publicity advertisement of the fifth issue of "Machine Learning High-end Training Camp" by Greedy Technology. The content of the course has been greatly updated compared with previous issues . On the one hand, new explanations on cutting-edge topics have been added as shown in the neural network (GCN, GAT, etc.) ), on the other hand, the core part (such as convex optimization, reinforcement learning) has increased the depth of the theoretical level. In addition, it will also include a series of topics such as scientific research methodology, meta-learning, interpretability, and Fair learning .
The following knowledge list may change with the number of subsequent issues, please feel free to search for updates on the search engine when learning.

Suitable for the crowd:

• Engaged in the AI ​​industry for many years, but I always feel that the technology is not deep enough, and I feel that I have encountered a technical bottleneck; 

• Staying on using models/tools, it is difficult to propose new models based on business scenarios ; 

• The optimization theory and cutting-edge technology behind machine learning are not deep enough;

• Plan to engage in cutting-edge scientific research, research work, and apply for graduate students and doctoral students in the field of AI; 

• Intend to enter top AI companies such as Google, Facebook, Amazon, Ali, etc.;

• Reading ICML, IJCAI and other conference articles is more difficult, seeming to understand but not understanding, unable to understand every detail;

------------------------------------------------------------------------------------------------------

01 Course Outline

 

Part 1: Convex optimization and machine learning

 

Week 1: Introduction to Convex Optimization

• Understanding machine learning from an optimization perspective

• The importance of optimization technology

• Common convex optimization problems

• Linear programming and Simplex Method

• Two-Stage LP

• Case: Explanation of transportation issues

 

Week 2: Explaining Convex Functions

• Convex Set Judgment

• First-Order Convexity

• Second-order Convexity

• Operations Preserve Convexity

• Quadratic Planning Problem (QP)

• Case: Least Squares Problem

• Project assignment: stock investment portfolio optimization

   

Week 3: Convex optimization problem

• Common convex optimization problem categories

• Semi-definite programming problem

• Geometric planning problem

• Optimization of non-convex functions

• Relaxation

• Integer Programming

• Case: Matching problem in taxi

 

Week 4: Duality

• Lagrangian dual function

• The geometric meaning of duality

• Weak and Strong Duality

• KKT conditions

• The dual problem of LP, QP, SDP

• Case: The dual derivation and realization of the classic model

• Other applications of duality

 

Week 5: Optimizing Technology

• First- and second-order optimization techniques

• Gradient Descent

• Subgradient Method

• Proximal Gradient Descent

• Projected Gradient Descent

• SGD and convergence

• Newton's Method

• Quasi-Newton's Method

 

 

Part 2 Graph Neural Network

 

Week 6: Fundamentals of Mathematics

• Vector space and graph theory foundation

• Inner Product, Hilbert Space

• Eigenfunctions, Eigenvalue

• Fourier transform

• Convolution operation

• Time Domain, Spectral Domain

• Laplacian, Graph Laplacian

 

Week 7: Graph Neural Networks in the Spectral Domain

• Convolutional neural network regression

• The mathematical meaning of convolution operation

• Graph Convolution

• Graph Filter

• ChebNet

• CayleyNet

• GCN

• Graph Pooling

• Case: Recommendation based on GCN

 

Week 8: Graph Neural Networks in the Spatial Domain

• Spatial Convolution

• Mixture Model Network (MoNet)

• Attention mechanism

• Graph Attention Network(GAT)

• Edge Convolution

• Comparison of spatial domain and spectral domain

• Project assignment: link prediction based on graph neural network

 

Week 9: Improvement and Application of Graph Neural Network

• Extension 1: Relative Position and Graph Neural Network

• Expansion 2: Incorporating Edge features: Edge GCN

• Extension 3: Graph Neural Network and Knowledge Graph: Knowledge GCN

• Extension 4: Gesture recognition: ST-GCN

• Case: Text classification based on graphs

• Case: Graph-based reading comprehension

 

 

Part Three Reinforcement Learning

 

Week 10: Basics of Reinforcement Learning

• Markov Decision Process

• Bellman Equation

• Three methods: Value, Policy, Model-Based

• Value-Based Approach: Q-learning

• Policy-Based Approach: SARSA

   

Week 11: Multi-A rmed Bandits

• Multi-Armed bandits

• Epsilon-Greedy

• Upper Confidence Bound (UCB)

• Contextual UCB

• LinUCB & Kernel UCB

• Case: Application case of Bandits in recommendation system

 

Week 12: Path planning

• Monte-Carlo Tree Search

• N-step learning

• Approximation

• Reward Shaping

• Combined with deep learning: Deep RL

• Project Assignment: Application Cases of Reinforcement Learning in Games

 

Week 13: RL in natural language processing

• Seq2seq model problem

• Custom loss combined with Evaluation Metric

• Custom loss combined with aspect

• Combination of different RL models and seq2seq models

• Case: RL-based text generation

 

 

Part Four Bayesian Method

 

Week 14: Introduction to Bayesian Methodology

• Bayes Theorem

• From MLE, MAP to Bayesian estimation

• Comparison of ensemble model and Bayesian method

• Computational Intractiblity

• Introduction to MCMC and Variational Method

• Bayesian linear regression

• Bayesian Neural Network

• Case: Named entity recognition based on Bayesian-LSTM

 

Week 15: Thematic model

• Generative model and discriminant model

• Hidden variable model

• The importance of Prior in Bayes

• Dirichlet distribution, polynomial distribution

• LDA generation process

• Parameters and hidden variables in LDA

• Supervised LDA

• Dynamic LDA

• Other variants of LDA

• Project assignment: Modify and build an unsupervised sentiment analysis model based on LDA

 

Week 16: MCMC method

• Detailed Balance

• Gibbs sampling for LDA

• Collapsed Gibbs sampling for LDA

• Metropolis Hasting

• Importance Sampling

• Rejection Sampling

• Large-scale distributed MCMC

• Big data and SGLD

• Case: LDA training based on distributed

 

Week 17: Variational Method

• Core Ideas of Variational Method

• Derivation of KL divergence and ELBo

• Mean-Field Variational Method

• EM algorithm

• Derivation of LDA by Variational Method

• Big data and SVI

• Comparison of Variational Method and MCMC

• Variational Autoencoder

• Probabilistic Programming

• Case: Using probabilistic programming tools to train Bayesian models

 

Week 18: Other cutting-edge topics

 

• Interpretability of the model

• Explain the CNN model

• Explain the sequence model

• Meta Learing

• Fair Learning

• Technology foresight