本人AI知識體系導航 - AI menu

 

Relevant Readable Links

 

Name	Interesting topic	Comment
Edwin Chen	非參貝葉斯
徐亦達老闆	Dirichlet Process	學習目標：Dirichlet Process, HDP, HDP-HMM, IBP, CRM
Alex Kendall	Geometry and Uncertainty in Deep Learning for Computer Vision	語義分割
colah's blog	Feature Visualization
Jason Yosinski	Understanding Neural Networks Through Deep Visualization
田淵棟		general CV
alexisbcook	Global Average Pooling Layers for Object Localization	目標定位
Tombone	http://www.computervisionblog.com/	DL, CV and the algorithms that are shaping the future of AI.

  

Others：

http://www.cnblogs.com/tornadomeet/archive/2012/06/24/2560261.html【理論總結挺好】 

http://www.cnblogs.com/charlotte77/【統計機器學習，可能實用】

http://blog.csdn.net/zhangjunhit/article/list/1【論文閱讀筆記不錯】

 

 

專一於數據分析之Kaggle and 圖像處理之AR on phone

 

---

 

How to be a Top AR Full-Stack Developer

 

正如該連接中所言，學習了哪些知識，計算機視覺纔算入門？

計算機視覺涉及面甚廣，找到一類問題好好研究並實踐就好，這類問題在本博客就指AR問題。

Ref: 計算機視覺入門書？

 

列出現代計算機視覺體系的主要科目（知識點）及其遞進關係。

一個單元表明一門course (12 weeks)或者一本book (600 pages)的學習量，親測。

按部就班很重要，後輩務必去掉大躍進的念頭。

人工智能之計算機視覺 - 學術體系
第四層	計算機視覺：模型，學習，推理
第三層	統計機器學習		深度學習
第二層	機器學習入門		計算機視覺入門
第一層	統計推斷	貝葉斯分析	多元線性分析	凸優化

 

編程是基本功，無須贅述。

人工智能之計算機視覺 - 軟件工程
第四層	實踐！實踐！實踐！
第三層	Android API, RN, OpenCV, Scikit-learning, ARToolkit, Unity
第二層	軟件架構，設計模式，代碼管理，單元測試
第一層	C/C++, Python, Java, Kotlin, Javascript, SQL

 

如上，乃基本的學習路線，僅是參考，仍可細分，但基本上具有了AR全棧開發者的潛力。

 

Phones with ARCore support, Feb, 2018

Indoor navigation app: you'll never be lost again

Inside Navigation【好東西，但時機不對】

 

---

 

My Hierarchy of AI Knowledge

實踐階段

若是你想要一個能走到冰箱面前而不撞到牆壁的機器人，那就使用 SLAM。

若是你想要一個能識別冰箱中各類物品的機器人，那就使用 Deep Learning。

基本上，這算一個風口；僅指路，不領路，需深耕。

 

加強現實 - Deep Learning 識別

綜述：

[Object Tracking] Overview of Object Tracking

[Object Tracking] Overview of algorithms for Object Tracking

 

輪廓識別：

[Object Tracking] Active contour model - Snake Model

[Object Tracking] Deep Boundary detection Tech

[Object Tracking] Contour Detection through Tensorflow running on smartphone

[Object Tracking] Contour Detection through OpenCV

 

目標定位：

[OpenCV] Real-time object detection with dnn module in OpenCV 3.3

[Localization] SSD - Single Shot MultiBoxDetector

[Localization] MobileNet with SSD

[Tensorflow] Android Meets TF in TensorFlow Dev Summit 2017

 

[Tensorflow] Object Detection API - prepare your training data

[Tensorflow] Object Detection API - build your training environment

[Tensorflow] Object Detection API - predict through your exclusive model

[Tensorflow] Object Detection API - retrain mobileNet

[Tensorflow] Object Detection API - mobileNet_v1.py

 

[Object Tracking] Identify and Track Specific Object

[Object Tracking] MeanShift

 

加強現實 - SLAM 跟蹤

[SLAM] 01. "Simultaneous Localization and Mapping"

[SLAM] 02. Some basic algorithms of 3D reconstruction

[SLAM] 03. ORB-SLAM2

[SLAM] AR Tracking based on which tools?

[ARCORE, Continue...]

 

 

 

衝刺階段

已看到收斂趨勢，查缺補漏，攻克難點疑點。

融會貫通方可運用自如，解決新問題。

  

生成式網絡 - Conv & Deconv

[Paper] Before GAN: sparse coding

Continue... 

 

深度學習概念 - UFLDL

[UFLDL] Basic Concept

[UFLDL] Linear Regression & Classification

[UFLDL] Dimensionality Reduction

[UFLDL] Generative Model

[UFLDL] Sparse Representation

[UFLDL] ConvNet

[UFLDL] Train and Optimize

 

深度學習理論 - Stats 385

[Stats385] Lecture 01-02, warm up with some questions

[Stats385] Lecture 03, Harmonic Analysis of Deep CNN

[Stats385] Lecture 04: Convnets from Probabilistic Perspective

[Stats385] Lecture 05: Avoid the curse of dimensionality

【暫時不實用，點到爲止】

 

統計機器學習 - PRML

9. [Bayesian] 「我是bayesian我怕誰」系列 - Gaussian Process

8. [Bayesian] 「我是bayesian我怕誰」系列 - Variational Autoencoders

7. [Bayesian] 「我是bayesian我怕誰」系列 - Boltzmann Distribution

6. [Bayesian] 「我是bayesian我怕誰」系列 - Markov and Hidden Markov Models

5. [Bayesian] 「我是bayesian我怕誰」系列 - Continuous Latent Variables

4. [Bayesian] 「我是bayesian我怕誰」系列 - Variational Inference

3. [Bayesian] 「我是bayesian我怕誰」系列 - Latent Variables

2. [Bayesian] 「我是bayesian我怕誰」系列 - Exact Inference

1. [Bayesian] 「我是bayesian我怕誰」系列 - Naive Bayes with Prior

  

 

  

混沌階段

打地基，處於強化學習初期的不穩定階段，感謝馬爾科夫收斂的性質，目標已收斂；自下向上，基本遵循按部就班的學習過程，夯實知識體系。

瞭解領域內的疑難點，認識技術細節的價值，爲下一階段作準備。

內容多爲早年整理，傾向於參考價值。

 

Bayesian Analysis

R與採樣方法：

[Bayes] What is Sampling

[Bayes] Point --> Line: Estimate "π" by R

[Bayes] Point --> Hist: Estimate "π" by R

[Bayes] qgamma & rgamma: Central Credible Interval

[Bayes] Hist & line: Reject Sampling and Importance Sampling

[Bayes] runif: Inversion Sampling

[Bayes] dchisq: Metropolis-Hastings Algorithm

[Bayes] prod: M-H: Independence Sampler for Posterior Sampling

[Bayes] Metroplis Algorithm --> Gibbs Sampling

[Bayes] Parameter estimation by Sampling

[Bayes] openBUGS: this is not the annoying bugs in programming

 

PGM基礎：

[PGM] What is Probabalistic Graphical Models

[PGM] Bayes Network and Conditional Independence

[PGM] Temporal Models

[PGM] Markov Networks

[PGM] Variable elimination and Belief propagation

[PGM] Exact Inference for calculating marginal distribution

 

貝葉斯基礎：

[BOOK] Applied Math and Machine Learning Basics

[Bayes] Multinomials and Dirichlet distribution

[Bayes] Understanding Bayes: A Look at the Likelihood

[Bayes] Understanding Bayes: Updating priors via the likelihood

[Bayes] Understanding Bayes: Visualization of the Bayes Factor

[Bayes] Why we prefer Gaussian Distribution

[Bayes] Improve HMM step by step

[Math] Unconstrained & Constrained Optimization

[Bayes] KL Divergence & Evidence Lower Bound

[Bayes] Variational Inference for Bayesian GMMs

[Bayes] Latent Gaussian Process Models

 

學習指南： 

[Math] A love of late toward Mathematics - how to learn it?

[Bayes ML] This is Bayesian Machine Learning 【原文總結得至關好】

 

 

Deep Learning

理論：

[BOOK] Applied Math and Machine Learning Basics             【DL書基礎，1至5章筆記】

[Hinton] Neural Networks for Machine Learning - Basic

[Hinton] Neural Networks for Machine Learning - Converage

[Hinton] Neural Networks for Machine Learning - RNN

[Hinton] Neural Networks for Machine Learning - Bayesian

[Hinton] Neural Networks for Machine Learning - Hopfield Nets and Boltzmann Machine

 

編程：

[Tensorflow] Architecture - Computational Graphs   【TF 框架】 

[Tensorflow] Practice - The Tensorflow Way     　　【相對基礎】

[Tensorflow] Cookbook - The Tensorflow Way 　　【前者的 Detail】

[Tensorflow] Cookbook - Neural Network         　　【代碼基礎寫法】

[Tensorflow] Cookbook - CNN                          　　【卷積網絡專題】

[Tensorflow] Cookbook - Object Classification based on CIFAR-10

[Tensorflow] Cookbook - Retraining Existing CNNs models - Inception Model

 

[Tensorflow] RNN - 01. Spam Prediction with BasicRNNCell

[Tensorflow] RNN - 02. Movie Review Sentiment Prediction with LSTM

[Tensorflow] RNN - 03. MultiRNNCell for Digit Prediction

[Tensorflow] RNN - 04. Work with CNN for Text Classification

 

[TensorBoard] Cookbook - Tensorboard

[TensorBoard] Train and Test accuracy simultaneous tracking

[TensorBoard] Name & Variable scope

 

訓練：

[Converge] Gradient Descent - Several solvers

[Converge] Weight Initialiser

[Converge] Backpropagation Algorithm 【BP實現細節】

[Converge] Feature Selection in training of Deep Learning 【特性相關性的影響】

[Converge] Training Neural Networks 【cs231n-lec5&6，推薦】

[Converge] Batch Normalisation

 

卷積：

[CNN] What is Convolutional Neural Network 【導論】

[CNN] Understanding Convolution 【圖像角度理解】

[CNN] Tool - Deep Visualization

 

模型：

[Model] LeNet-5 by Keras

[Model] AlexNet

[Model] VGG16

[Model] GoogLeNet

[Model] ResNet　　

[Localization] R-CNN series for Localization and Detection

[Localization] YOLO: Real-Time Object Detection

[Localization] SSD - Single Shot MultiBoxDetector

[Localization] MobileNet with SSD

 

其餘：

[GPU] CUDA for Deep Learning, why?

[GPU] DIY for Deep Learning Workstation

[Keras] Install and environment setting

[Keras] Develop Neural Network With Keras Step-By-Step 

[GAN] *What is Generative networks 【導論，」生成式模型「有哪些，與」判別式模型「同級】

[GAN] How to use GAN - Meow Generator 

[DQN] What is Deep Reinforcement Learning 【導論：此方向優先級低】

[Understanding] Compressive Sensing and Deep Model 【感知壓縮，暫且不懂】

[DL] *Deep Learning for Industry - Wang Yi 【課外閱讀】

  

 

Machine Learning

  

/* ML文件夾待整理 */

 

 

IR & NLP基礎

檢索：

[IR] Boolean retrieval

[IR] Index Construction

[IR] Compression

[IR] Tolerant Retrieval & Spelling Correction & Language Model

[IR] Probabilistic Model

[IR] Link Analysis

[IR] Ranking - top k

[IR] Evaluation

[IR] Information Extraction

[IR] Open Source Search Engines

[IR] Search Server - Sphinx

[IR] Concept Search and LSI

[IR] Concept Search and PLSA

[IR] Concept Search and LDA

 

壓縮：

[IR] What is XML

[IR] XML Compression

[IR] Advanced XML Compression - ISX

[IR] Advanced XML Compression - XBW

[IR] XPath for Search Query

[IR] Graph Compression

[IR] Bigtable: A Distributed Storage System for Semi-Structured Data

[IR] Huffman Coding

[IR] Arithmetic Coding

[IR] Dictionary Coding

[IR] BWT+MTF+AC

[IR] String Matching

[IR] Suffix Trees and Suffix Arrays

[IR] Time and Space Efficiencies Analysis of Full-Text Index Techniques

[IR] Extraction-based Text Summarization

 

其餘：

[IR] Word Embeddings

 

【以上內容需隨recommended system一塊兒再過一遍，完善體系】 

 

 

AR基礎

[Artoolkit] Marker Training

[Artoolkit] ARToolKit's SDK Structure on Android

[Artoolkit] Framework Analysis of nftSimple

[Artoolkit] kpmMatching & Tracking of nftSimple

[Artoolkit] Android Sample of nftSimple

[Artoolkit] Can I Use LGPL code for commercial application

[Artoolkit] Marker of nftSimple

[Artoolkit] ARSimpleNativeCarsProj for Multi Markers Tracking

 

[Unity3D] 01 - Try Unity3D

[Unity3D] 02 - ** Editor Scripting, Community Posts, Project Architecture

[Unity3D] 03 - Component of UI

[Unity3D] 04 - Event Manager

[Unity3D] 05 - Access to DB or AWS

【簡單涉及3D建模知識點，非重點】

 

　　  

CV基礎

概念：

[OpenCV] Install openCV in Qt Creator

[OpenCV] Basic data types - Matrix

[OpenCV] IplImage and Operation

[OpenCV] HighGUI

[OpenCV] Image Processing - Image Elementary Knowledge

[OpenCV] Image Processing - Grayscale Transform

[OpenCV] Image Processing - Frequency Domain Filtering

[OpenCV] Image Processing - Spatial Filtering

[OpenCV] Image Processing - Fuzzy Set

[OpenCV] Feature Extraction

[OpenCV] Feature Matching

[SLAM] Little about SLAM

[SLAM] Camera math knowledge

[Tango] Basic Knowledge

 

實踐：

// 內容將合併，從新整理

[OpenCV] Samples 01: drawing【幾何圖案、文字等】

[OpenCV] Samples 02: [ML] kmeans【聚類算法】

[OpenCV] Samples 03: cout_mat【Mat計算能力】

[OpenCV] Samples 04: contours2【二值圖案找輪廓】

[OpenCV] Samples 05: convexhull【散點的凸包輪廓】

[OpenCV] Samples 06: [ML] logistic regression【線性二分類】

[OpenCV] Samples 07: create_mask【鼠標圈圖】

[OpenCV] Samples 08: edge【邊緣檢測】

[OpenCV] Samples 09: plImage <==> Mat 【色域通道分離】

[OpenCV] Samples 10: imagelist_creator【圖片地址list參數】

[OpenCV] Samples 11: image sequence【視頻流提取】

[OpenCV] Samples 12: laplace【視頻流處理】

[OpenCV] Samples 13: opencv_version【版本信息顯示】

[OpenCV] Samples 14: kalman filter【預測下一個狀態】

[OpenCV] Samples 15: Background Subtraction and Gaussian mixture models【背景差分】

[OpenCV] Samples 16: Decompose and Analyse RGB channels【色域通道分離】

[OpenCV] Samples 17: Floodfill【聚類算法】

[OpenCV] Samples 18: Load image and check its attributes【圖片屬性】

 

擴展： 

[CNN] Face Detection

[Android Studio] Using Java to call OpenCV

[Android Studio] Using NDK to call OpenCV

[OpenCV] Install OpenCV 3.3 with DNN

[OpenCV] Install OpenCV 3.4 with DNN

 

趣碼收集：

[Link] Face Swap Collection 

[Link] Face Swap without DLIB【代碼可用】

 

  

算法基礎

[Algorithm] Deferred Acceptance Algorithm

[Algorithm] Beating the Binary Search algorithm – Interpolation Search, Galloping Search

[Algorithm] Warm-up puzzles

[Algorithm] Asymptotic Growth Rate

[Algorithm] Polynomial and FFT

[Algorithm] Maximum Flow

[Algorithm] String Matching and Hashing

[Optimization] Greedy method

[Optimization] Dynamic programming

[Optimization] Advanced Dynamic programming

  

---

Everything here starts from 2016