This article is an introduction to all published articles and will be updated frequently in the future.
Table of contents
1. Data mining, machine learning, deep learning, cloud computing, artificial intelligence
2. Deep learning, reinforcement learning, confrontation learning, transfer learning
3. Basic knowledge - linear algebra
4. Basic knowledge--probability and mathematical statistics
6.1 What is training and what is inference?
6.2 General Process of Machine Learning
6.3 Machine Learning Problem Types
7.1 Feedforward Neural Networks
7.2 Convolutional Neural Networks
1. Data mining, machine learning, deep learning, cloud computing, artificial intelligence
- Data mining: "mining" hidden information from massive data
- Machine learning: feeding data into a computer program so that the computer learns new knowledge
- Deep Learning: Automatically Learning Features from Big Data
- Cloud Computing: Large-Scale Distributed IT Infrastructure
- Artificial intelligence: a high degree of consensus on the replacement of human beings based on learning and cognitive composition
The relationship between them:
Data is the source of artificial intelligence. Technologies such as big data parallel computing and stream computing are the guarantee for the practical application of artificial intelligence. Artificial intelligence is the main method of big data, especially complex data analysis.
Machine Learning Basics: Machine Learning (1) - Basic Concepts_三三木木七的博客-CSDN Blog .
2. Deep learning, reinforcement learning, confrontation learning, transfer learning
(To be added)
3. Basic knowledge - linear algebra
3.1 Scalars, vectors, matrices, tensors
3.2 Matrix and vector multiplication
3.3 Identity matrix and inverse matrix
3.4 Linear correlation and generating subspace
3.5 Norm
3.6 Orthogonal vectors and subspaces
3.7 Special matrices and vectors
3.8 Eigen Decomposition
3.9 Singular value decomposition
3.10 MP pseudo-inverse
4. Basic knowledge--probability and mathematical statistics
4.1 Probability, probability spaces, random variables
4.2 Probability distribution
4.3 Conditional probability
4.4 Chain Rule, Bayes Theorem
4.5 Mathematical expectations
4.6 Variance
4.7 Correlation coefficient
4.8 Maximum Likelihood Estimation
5. Common tool library
5.1 Numpy
Numpy, which provides the basic library for scientific computing, mainly provides N-dimensional array implementation, computing power and basic mathematical algorithms. (Basic Scientific Computing Library)
Details: (1 message) Machine Learning (2)--NumPy
5.1.1 Tensors ✅
element-by-element calculation
matrix addition
scalar and matrix multiplication
relu operation
5.1.2 Broadcast
5.1.3 Tensor dot product
(1) Vector dot product
same number of vector elements
(2) Convolution
5.1.4 Matrix multiplication
5.1.5 Tensor Transformation
Change the rows and columns of the tensor to get the desired shape
The total number of elements of the transformed tensor is the same as the original tensor
A special tensor transformation -- transpose: row and column swapping
5.1.6 Downsampling
5.2 Scipy
Scipy, based on a series of mathematical operation methods developed by Numpy. Data manipulation, parallel computing, etc. (powerful set of scientific computing tools)
5.3 Pandas
Pandas, a powerful time series dataset processing tool. Depends on Numpy. (A sharp tool for data analysis)
5.4 Matplotlib
Matplotlib, a data visualization toolkit
5.5 Scikit-learn
Scikit-learn, machine learning
6. Machine Learning
6.1 What is training and what is inference?
Machine Learning: People input data and the answers expected from that data, and the system outputs rules (training).
Prediction/Inference: These rules are then applied to new data and allow the computer to generate answers autonomously.
6.2 General Process of Machine Learning
In a practical sense, machine learning is a method of using data, training a model, and then using the model to predict .
General process: data->model->loss function->optimization->tuning->result analysis->algorithm extension->deployment and operation->infrastructure
Data: experience storage
Model: Select the base model based on the problem and data
Loss function: A measure of model accuracy
Optimization: eg gradient descent method minimizes loss function
Tuning: Tuning model hyperparameters
Analysis of Results: Comparing Model Accuracy vs Run Time
Algorithm Extensions: Validation Set Tests
Deployment and operation: real-time, stable, and expanded
Infrastructure: inference scale, interface specification, computing power
6.3 Machine Learning Problem Types
Classification
return
clustering
abnormal detection
reinforcement learning
7. Deep Learning
Deep learning is a type of machine learning model that combines low-level features to form more abstract high-level representation attribute categories or features to discover feature representations of data .
7.1 Feedforward Neural Networks
Input layer, hidden layer, output layer.
Neurons start from the input layer, accept the input from the previous level, and output to the next level, until the output layer.
There is no feedback in the entire network, which can be represented by a directed acyclic graph.
7.2 Convolutional Neural Networks
Multi-layer neural network combines local connection, weight sharing and sub-sampling to extract image/text features.
7.3 Recurrent Neural Networks
A neural network that models sequence data, where the current output of a sequence is also related to the previous output.
7.4 Transfer Learning
Migration learning is to transfer the model parameters that are easy to train to a new model to help the new model training. The model parameters that have been learned can be shared with the new model in some way, so as to speed up and optimize the learning efficiency of the model.
7.5 Adversarial Learning
The core of the generative confrontation network is confrontation, two networks compete with each other, one is responsible for generating samples, and the other is responsible for discriminating samples
7.6 Reinforcement Learning
Reinforcement learning studies the task of interaction between the agent and the environment, and learns how to map the environment to actions to obtain the largest, numerical, reward signal.