Amazing GANs

  • R2
  • Day 3, 13:00‑13:45
  • Chinese talk w. English slides
  • Science
  • Intermediate

Unsupervised learning is how human learns. And generative adversarial networks is one of the most popular algorithms to do unsupervised learning. There are lots of GAN papers with amazing results. I would like to talk about issues of replicating GAN results and how I solved them.

Talk Detail

There are many papers try to build models to generate images. Some of them try to tackle the problem from different point of views (probability or energy). Trainings fail easily if bad parameters are used. I want to talk about how the parameters (e.g. large learning rate) affect the trainings.

Speaker Information


覺得有趣選了物理系,結果差點畢不了業。想做遊戲所以開始寫程式,結果沒進過遊戲公司。覺得電腦圖學好玩,結果接觸了電腦視覺。覺得機器學習很酷,結果待續。現在在 KKStream 披著 Android Adaptive Player 的皮,企圖跟同事強迫推銷自己都不太懂的演算法。