Efficient Net

  • 현재 트렌드로는 다양한 합성곱 신경방등이 넓이(width), 깊이(depth), 해상도(resolution) 등의 여러가지 다양한 특성으로 합성곱 신경망 모델을 구성
    • depth – resnet같은 경우를 예로 들면, 깊이에 따라 정확도가 높아짐 (일정이상은 높아지지 않고, 1000이상이면 무용지물)
    • width – 여러가지 filter를 적용하여 width가 길어진 네트워크는 많은 fine-grain feature를 훈련할수 있음. 다만 너무 깊이가 얕으면 추상적레벨의 feature를 반영하지 못할 수는 있음.
    • resolution – 당연히 해상도가 높아지면 정확도가 높아지지만, 600X600정도 이상이면 정확도 상승폭에 비해 하드웨어 무리
  • 결국 정확도 상승을 위해 극한의 하드웨어 성능이 필요한 정도로 모델이 커지지만, 상승폭은 계속 줄어들게 됨.
  • 이 셋의 밸런스를 잘 맞추는것이 제일 가성비(?)가 좋은 네트워크를 만들게 됨.
  • 따라서 Efficient Net은 메모리, CPU에 따라 제일 좋은 효과를 얻을 수있는 공식을 찾아내기 시작
  • 효과는 다음과 같이, 적은 자원으로도 State of the art 를 뛰어넘는 높은 정확도를 달성할수 있음을 보임

9 thoughts on “Efficient Net

  1. Hello there, just became alert to your blog through Google,
    and found that it is truly informative. I’m gonna watch
    out for brussels. I will appreciate if you continue this
    in future. Lots of people will be benefited from your writing.
    Cheers!

    Here is my blog post: thing g

  2. Hey there! I’ve been reading your site for some time now and finally got the
    courage to go ahead and give you a shout out from Huffman Tx!
    Just wanted to say keep up the great work!

    My website: rsacwgxy g

  3. I love your blog.. very nice colors & theme.

    Did you design this website yourself or did you hire
    someone to do it for you? Plz respond as I’m looking to construct my own blog and would like to know where u got this from.
    appreciate it

    My web blog :: cbd oil; tinyurl.com,

  4. It’s awesome to visit this site and reading the views of all friends
    regarding this paragraph, while I am also keen of getting experience.

    Take a look at my web blog :: cbd oil (separ.es)

  5. Write more, thats all I have to say. Literally, it seems
    as though you relied on the video to make your point.
    You clearly know what youre talking about, why throw away your intelligence on just posting videos to your site when you could be giving us something enlightening to read?

    my homepage: cbd oil that works 2020

댓글 남기기