Guide line of AI
Download as PPTX, PDF0 likes56 views
This document discusses different types of artificial intelligence including supervised learning, unsupervised learning, semi-supervised learning, transfer learning, and reinforcement learning. It provides examples of each type such as using supervised learning for PM2.5 prediction and income prediction, unsupervised learning for image clustering, semi-supervised learning for self-training and 3D data augmentation, transfer learning for medical imaging models, and reinforcement learning for playing Atari games. It also discusses interesting topics in AI like designing AI chips, data labeling, and producing high quality labeled datasets.
More Related Content
Similar to Guide line of AI (20)
Recently uploaded (20)
Guide line of AI
- 1. \ AI Tsun Hsien Tang Data science program in NTU NLP lab 1
- 2. Outline ¢ What is AI ¢ Scope in AI ＄ Supervised learning (PM2.5, Income porediction) ＄ Semi-Supervised learning (self-learning, 3D data augmentation) ＄ Unsupervised learning (Image Clustering) ＄ Transfer learning (ChexNet) ＄ Reinforcement learning (OpenAI gym: atari game) ¢ Interesting topics in AI ＄ Design Automation of Deep Learning Systems (AI chip, model compression) ＄ Data labeling 2
- 4. What is AI Neural network AlphaGo Self driving 4
- 5. What is AI 廨屹薑腮圜 徭 隼 Z 冱 尖 廨勗宥 Y垢殻 嶄猟秘隈 Markov chain, Viterbi algorithm, Language model, Corpus Google Map shortest path planning AI Course 5
- 6. Scope in Machine Learning 6 ¢ Image synthesis f ( ) = f ( ＾black hair, blue eyes￣, noise ) = http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2017_2/Lecture/introduction.pdf
- 7. Scope in Machine Learning http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2017_2/Lecture/introduction.pdf 7
- 8. Supervised learning ￣Right answer given￣ 8 Training data PM2.5 prediction (regression)
- 9. Supervised learning KDD cup 2018:AO隆 48 弌r坪 PM2.5, PM10 才 O3議盒 9 ￣Right answer given￣
- 10. Supervised learning 10 ￣Right answer given￣ Income prediction (classification) Determine whether a person makes over 50K a year. Training data Ground truth Text Sentiment Classification (classification) Positive or Negative
- 11. Supervised learning 11 Feature importance ￣Right answer given￣ Algorithm: SVM Random forest Descicion tree Logistic regression Linear regression Neural network Gradient boosting tree Microsoft LightGBM
- 12. Unsupervised learning ￣No right answer given￣ 12 Image Clustering mnist fashion mnist Algorithm: K-means, Hierarchical clustering, DBSCAN feature extraction + clustering algo
- 14. Semi-Supervised learning 14 3D data augmentation Learning Deep Object Detectors from 3D Models, Peng et al., ICCV 2016
- 15. Transfer learning 15 http://slideplayer.com/slide/8370683 / Diseases prediciton ImageNet Chest X-ray Database
- 17. Interesting topics in AI AI chip & model compression 17 Autonomous vehicles object detection Quadcopter people tracking
- 18. Interesting topics in AI ¢ Data labeling ¢ 繁垢崘斯: 嗤謹富繁垢嘉嗤謹富崘嬬 ¢ Most of the well-known datasets are published by Microsoft, Google or well- financed Lab. 18 岬街と勃濃W丕: Produce high quality data labeling Reliability and Diversity
- 19. About Kaggle ¢ Feature Engineering ＄ Rely on experience and programming skill ¢ Algorithms ＄ Find the state-of-the-art related work ＄ XXXX github ＄ XGBoost ＄ Keras ¢ Ensembling ¢ Experiment record ¢ Luck 19