Conditional Variational AutoEncoder (Cond_VAE) - Example using MNIST dataset
In this video, I showed how a Conditional Variational AutoEncoder (𝐂𝐨𝐧𝐝_𝐕𝐀𝐄) can be implemented using the 𝐌𝐍𝐈𝐒𝐓 dataset.
Conditional Variational AutoEncoder (𝐂𝐨𝐧𝐝_𝐕𝐀𝐄) is a generative model which assumes that the data is generated by some random process, involving unobserved continuous random variables. Like any other AutoEncoder, it has an 𝗲𝗻𝗰𝗼𝗱𝗲𝗿 and a 𝗱𝗲𝗰𝗼𝗱𝗲𝗿.
The 𝗲𝗻𝗰𝗼𝗱𝗲𝗿 tries to learn q(z|x, y) which is the same as learning the hidden representation of the data x, conditioned on y (in this case, the labels of 𝐌𝐍𝐈𝐒𝐓 images). The 𝗱𝗲𝗰𝗼𝗱𝗲𝗿 on the other hand tries to learn p(x|z, y) which is just decoding the hidden representation to input space conditioned by y.
𝑮𝒊𝒕𝑯𝒖𝒃 𝒂𝒅𝒅𝒓𝒆𝒔𝒔: https://github.com/randomaccess2023/M...
𝙆𝙇 𝘿𝙞𝙫𝙚𝙧𝙜𝙚𝙣𝙘𝙚 𝙛𝙤𝙧 𝙑𝘼𝙀𝙨: https://stats.stackexchange.com/quest...
00:48 - 𝘐𝘮𝘱𝘰𝘳𝘵 𝘳𝘦𝘲𝘶𝘪𝘳𝘦𝘥 𝘭𝘪𝘣𝘳𝘢𝘳𝘪𝘦𝘴
03:38 - 𝘋𝘦𝘧𝘪𝘯𝘦 𝘵𝘩𝘦 𝘥𝘦𝘷𝘪𝘤𝘦
05:13 - 𝘓𝘰𝘢𝘥 𝙈𝙉𝙄𝙎𝙏 𝘥𝘢𝘵𝘢𝘴𝘦𝘵
11:45 - 𝘋𝘦𝘧𝘪𝘯𝘦 𝘥𝘢𝘵𝘢𝘭𝘰𝘢𝘥𝘦𝘳𝘴
13:55 - 𝘋𝘦𝘧𝘪𝘯𝘦 𝘵𝘩𝘦 𝘯𝘦𝘵𝘸𝘰𝘳𝘬 (𝘾𝙤𝙣𝙙_𝙑𝘼𝙀) 𝘢𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘶𝘳𝘦
26:57 - 𝘝𝘪𝘴𝘶𝘢𝘭𝘪𝘻𝘦 𝘵𝘩𝘦 𝘯𝘦𝘵𝘸𝘰𝘳𝘬 (𝘾𝙤𝙣𝙙_𝙑𝘼𝙀) 𝘢𝘳𝘤𝘩𝘪𝘵𝘦𝘤𝘵𝘶𝘳𝘦
30:16 - 𝘋𝘦𝘧𝘪𝘯𝘦 𝘵𝘩𝘦 𝘵𝘳𝘢𝘪𝘯𝘪𝘯𝘨 𝘣𝘢𝘵𝘤𝘩
32:13 - 𝘋𝘦𝘧𝘪𝘯𝘦 𝘵𝘩𝘦 𝘵𝘦𝘴𝘵 𝘣𝘢𝘵𝘤𝘩
33:49 - 𝘚𝘦𝘵 𝘶𝘱 𝘮𝘰𝘥𝘦𝘭 (𝘾𝙤𝙣𝙙_𝙑𝘼𝙀), 𝘤𝘳𝘪𝘵𝘦𝘳𝘪𝘰𝘯 (𝙡𝙤𝙨𝙨_𝙛𝙪𝙣𝙘𝙩𝙞𝙤𝙣) 𝘢𝘯𝘥 𝘰𝘱𝘵𝘪𝘮𝘪𝘻𝘦𝘳
36:22 - 𝘛𝘳𝘢𝘪𝘯 𝘾𝙤𝙣𝙙_𝙑𝘼𝙀
43:32 - 𝘗𝘭𝘰𝘵 𝘭𝘰𝘴𝘴𝘦𝘴
46:09 - 𝘔𝘰𝘥𝘦𝘭 𝘱𝘳𝘦𝘥𝘪𝘤𝘵𝘪𝘰𝘯
52:53 - 𝘊𝘳𝘦𝘢𝘵𝘦 𝘪𝘮𝘢𝘨𝘦𝘴 𝘧𝘳𝘰𝘮 𝘯𝘰𝘪𝘴𝘦 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯𝘦𝘥 𝘰𝘯 𝘭𝘢𝘣𝘦𝘭𝘴
#generativemodels #datascience #deeplearning #artificialintelligence #neuralnetworks #ConditionalVAE #autoencodermodels