content
1 Basic theory
1.1 Basic Concepts
(1) Neural network
You can refer to learning [Neural Network Classification]
(2) Perceptron
(3) Gradient descent
(4) Multilayer Perceptron
(5) Backpropagation
(6) Loss function
(7) Optimizer
(8) Convolution
(9) Pooling
(10) Attention Attention
。。。
(11) Three major application areas
- Graphics processing CV: OCR, object detection, image recognition, etc. Commonly used networks are LetNet, VGGNet, GoolgNet, ResNet, GAN, improved Transformer, etc.
- Natural language processing NLP: word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, keyword extraction, text classification, automatic summarization and information retrieval, etc. Commonly used networks are TextCNN, TextRNN, Transformer, Bert.
- Speech Recognition ASR: Speech Synthesis and Speech Recognition. Commonly used networks such as CNN and improved CNN.
Note: In different fields, there are different concepts that need to be systematically learned. Click here to stop, and follow-up will provide detailed guide articles for each field, such as pixel, resolution, sampling, texture, etc. in image processing. In natural language processing, there are word vectors, word embeddings, etc.
Recommended books: "Machine Learning - Zhou Zhihua" "Deep Learning - Flower Book"
Recommended online courses: "Machine Learning - Wu Enda" "Machine Learning - Li Hongyi"
Recommended study notes: github has the most Star notes
1.2 Languages and tools
Programming language: Python
Compiler: VScode or pycharm
Python needs to be learned systematically and used to understand the use of some grammars, so that in the subsequent process, you can easily read other people's source code and write concise and efficient programs. As a general-purpose compiler, VScode is written in most languages. It is ecologically sound, easy to debug, easy to run, and easy to deploy. Among them, it is very necessary to master the code writing and running of the notebook's ipynb file, which is a file format that combines Markdown syntax with image visualization. It can be edited and used directly in VScode. After downloading the installation package on the official website, install it directly with one click, and then configure the Python compilation environment according to the tutorial to use it. The use of Vscode can play a variety of flowers, and if you are interested, search for "the strongest Vscode". Configure VScode to be the most arrogant and most NB compiler.
Recommended books: "Stupid Way to Learn Python" "Liao Xuefeng's Python Tutorial"
Recommended online course: small turtle python tutorial
VScode Python Environment Configuration: Configuration Tutorial
1.3 Deep Learning Framework
（1）Pytorch
（2）TensorFlow2.0
Recommended books: "Hands-on Deep Learning-pytorch", "Pytorch Deep Learning Introduction and Practice", "TensorFlow 2.0 Deep Learning from Scratch", "Simple and Rough TensorFlow 2.0", "Eat That TensorFlow2.0 in 30 Days"
Recommended website: Pytorch Chinese official website , Pytorch deep learning actual combat-B station , Peking University-Tensorflow2.0
2 Project combat
Determine which field you are working on, try a few more projects in this field, and recommend the book "Alibaba Cloud Tianchi Competition Question Analysis - Deep Learning", which has four deep learning examples, very detailed explanations, and open source. Examples of other practical projects in the three aspects are as follows
2.1 Image Processing
(1) PyTorch implements image style transfer
Standard star: 3.1k+
Address: [github address]
(2) StarGan: PyTorch implementation of multi-domain image translation GAN network
Standard star: 2.1k+
Address: [github address]
(3) Basic tutorial on the combination of TensorFlow and OpenCV, introduction to artificial intelligence image processing
Stars: 30+
Address: [github address]
2.2 Natural Language Processing
(1) Introductory exercises for natural language processing
Stars: 964+
Address: [github address]
(2) Tianchi-News Text Classification Competition
Summary of multiple open source solutions
[BetterBench - Summary of Multiple Solutions for News Text Classification]
2.3 Speech Recognition
(1) Voice digital recognition
Star: 5+
Address: [github address]
(2) Chinese speech recognition system based on deep learning
Standard star: 5.2k+
Address: [github address]
(3) Mandarin recognition based on deep learning
Stars: 4+
Address: [github address]
3 Competition Advanced
Participate in the competition to play the rankings, ranging from a month to half a year. Participating once will benefit you for a lifetime. After two complete competitions, you can complete the entry and become an advanced player.
(1) Aliyun Tianchi
(2) Hewhale Community
(3) Huawei Cloud
(4) Baidu Flying Paddle
(5) Tencent Cloud
(6) FlyAI competition
(7) Biendata competition
(8) DataFountain competition
(9) DataCastle competition
(10) Kaggle competition