广东工业大学：《机器学习》课程教学资源（课件讲义）第21讲 生成式网络模型（VAE Generation）

Unsupervised Learning: Generation

Unsupervised Learning: Generation

http://www.rb139.com/index.ph p?s=/Lot/44547 Creation Drawing? 唐三 唐三 唐三百 Writing Poems?

Creation Drawing? Writing Poems? http://www.rb139.com/index.ph p?s=/Lot/44547

Creation 。Generative Models: https://openai.com/blog/generative-models/ Wht0 eot onato 0 ko mot undenmind 气G w人ow衣入o2 2-0 Hale What I cannot create, I do not understand. ®三， 9t)a0 Richard Feynman @1=1Y以以④ Copyright California Institute of Technology.All rights reserved. Commercial use or modification of this material is prohibited. https://www.quora.com/What-did-Richard-Feynman-mean-when-he-said-What-l- cannot-create-l-do-not-understand

Creation • Generative Models: https://openai.com/blog/generative-models/ https://www.quora.com/What-did-Richard-Feynman-mean-when-he-said-What-I￾cannot-create-I-do-not-understand What I cannot create, I do not understand. Richard Feynman

Creation Now V.S. In the future Machine draws a cat http://www.wikihow.com/Draw-a-Cat-Face

Creation Now In the future v.s. Machine draws a cat http://www.wikihow.com/Draw-a-Cat-Face

Generative Models Component-by-component Autoencoder Generative Adversarial Network (GAN)

Generative Models Component-by-component Autoencoder Generative Adversarial Network (GAN)

Component-by-component 。Image generation NN →■ E.g.3 x 3 images NN →口 NN : : Can be trained just with a large collection of images without any annotation

Component-by-component • Image generation E.g. 3 x 3 images NN NN NN…… Can be trained just with a large collection of images without any annotation

Component-by-component 。Image generation NN E.g.3 x 3 images NN →口 NN Can be trained just with a large collection of images without any annotation

Component-by-component • Image generation E.g. 3 x 3 images NN NN NN…… Can be trained just with a large collection of images without any annotation

Practicing Generation Models: Pokemon Creation Small images of 792 Pokemon's Can machine learn to create new Pokemons? Don't catch them!Create them! 。Source of image: http://bulbapedia.bulbagarden.net/wiki/List_of_Pok%C3%A 9mon_by_base_stats_(Generation_VI) Original image is 40 x 40 Making them into 20 x 20

Practicing Generation Models: Pokémon Creation • Small images of 792 Pokémon's • Can machine learn to create new Pokémons? • Source of image: http://bulbapedia.bulbagarden.net/wiki/List_of_Pok%C3%A 9mon_by_base_stats_(Generation_VI) Don't catch them! Create them! Original image is 40 x 40 Making them into 20 x 20

Practicing Generation Models: Pokemon Creation ·Tips(R) >Each pixel is represented by 3 numbers(corresponding to RGB) R=50,G=150,B=100 >Each pixel is represented by a 1-of-N encoding feature Clustering the similar color 167 colors in total

Practicing Generation Models: Pokémon Creation • Tips (?) R=50, G=150, B=100 ➢Each pixel is represented by 3 numbers (corresponding to RGB) ➢Each pixel is represented by a 1-of-N encoding feature 0 0 1 0 0 …… Clustering the similar color 167 colors in total

Practicing Generation Models: Pokemon Creation Original image(40 x 40): http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_creation/ima ge.rar ·Pixels(20x20: http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_creation/pixe I color.txt Each line corresponds to an image,and each number corresponds to a pixel http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_cre ation/colormap.txt →255255255 0 ,535353 0000000000000000000000 7494949 000194134001900000000000 186186186 0000120000000000000000 515151 000000014474445100000000 545454 000000012180808100000000 187187187 00000123183522052000000 838383 93949393859538969798999967999 0000001106106106106106611070 505152 251251251 525252 Following experiment:1-layer LSTM,512 cells

Practicing Generation Models: Pokémon Creation • Original image (40 x 40): http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_creation/ima ge.rar • Pixels (20 x 20): http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_creation/pixe l_color.txt • Each line corresponds to an image, and each number corresponds to a pixel • http://speech.ee.ntu.edu.tw/~tlkagk/courses/ML_2016/Pokemon_cre ation/colormap.txt • Following experiment: 1-layer LSTM, 512 cells …… … 0 1 2