Haitham Bou Ammar
RL team leader @Huawei R&D UK & UCL H. Assistant Prof.
发布于
评论(3)
  • Double Q
    Double Q 回复
    Thank you, many of your pictures have helped me understand a lot, this work is very meaningful!
  • Haitham Bou Ammar
    Haitham Bou Ammar 回复
    oh so the log trick is like this: gradient of log f(x) = f'(x)/f(x) . This mean that f'(x) = f(x) log f(x). So that will allow us to replace f(x) log f(x) as f'(x) and go forward.
  • Ai 回复
    What does the log-trick used in step 3 mean?