摘要：
There are some problems: mismatch of model and reality; gradient explosion so, the dynamics can be quite messy, and backpropogating can be quite probl 阅读全文
摘要：
... 阅读全文
摘要：
solved normally by sequential quadratic programming algorithms an example of linear system 阅读全文
摘要：
You have to force experts to treat some uncommon and extreme situations. a mechanical way to learn However, we don't know rt if you use sequence GAN, 阅读全文
摘要：
not only JS divergence could be applied to GAN, other divergences are all applicable! f start is convex several ACG icons become very similar, if trai 阅读全文
摘要：
Too much limitation of Gaussian model. The images are too blurry. So any general model? But if PG(x;θ) is a neural network, it's impossible to calcula 阅读全文
摘要：
HW2: input a sentence, output an ACG icon 3 target: trains from front view, side views. So that the output would be the average of the three pictures. 阅读全文
摘要：
比较有用的是conditioned generator，能够控制输入的vector来控制对应的文字音像 https://zhuanlan.zhihu.com/p/24767059 单纯生成人脸意义不大，因为随便拍一个路人就行了。 但是能从左右照片生成正面照片，就很神奇了 必须学会辨别转折 Varia 阅读全文
摘要：
https://www.bilibili.com/video/av15997678/ My own deep reinforcement learning code: https://github.com/ysgclight/Reinforcement-Learning-with-Pytorch D 阅读全文
摘要：
data augumentation 阅读全文
摘要：
10 free hours run on AWS click this one click on new machine pick a region choose linux ubuntu 16 250GB is preferred ctrl shift v to paste your passwo 阅读全文
摘要：
https://www.bilibili.com/video/av22940029 left hand side: NN being constructed right hand side: NN being called turn the NN code into GPU compatible m 阅读全文
摘要：
high bias if the robot has learnt something (no changes appear with iterations) however, in the real world tasks, the task could change a little bit, 阅读全文
摘要：
model free: high variance. model based: high bias within 1h of human demonstration of each task, VR!!! 阅读全文
摘要：
intrinsic ambiguity: move toward purple triangle? move away from red triangle? move along grey arrow? or the combine of them? the right part of the ri 阅读全文
摘要：
阅读全文
摘要：
So, the process is similar to one-to-many RNN? learn much more efficiently than model-free method iteratively get better less than 300 trials ~ 25min 阅读全文
摘要：
you wouldn't try to explore any problem structure in DFO low dimension policy 30 degrees of freedom 120 paramaters to tune keep the positive results i 阅读全文
摘要：
^ is the square root of epsilon a simplified version of hard version a more smooth way to find correct solution the first term is the REINFORCE term, 阅读全文
摘要：
fast feedback to robot with better shape reward func, and learning could be much faster open ai baseline rllab multiple tasks and multiple seeds to te 阅读全文