The data is more important than the algorithm in machine learning

Поделиться
HTML-код
  • Опубликовано: 7 ноя 2024

Комментарии • 11

  • @matthewwithum8372
    @matthewwithum8372 3 года назад +3

    Beautiful philosophy.

  • @yiyangwu1314
    @yiyangwu1314 3 года назад +2

    Parameters datas input, is dynamically activating

  • @Dynamite_mohit
    @Dynamite_mohit 3 года назад +3

    I wish lex could know that I exist.
    He is like a big brother, I always needed.

  • @jcalbi123
    @jcalbi123 3 года назад +1

    Ocean Protocol $ocean

  • @rajbanwait325
    @rajbanwait325 3 года назад +6

    bias in data, huge problem

  • @davidregi7571
    @davidregi7571 3 года назад

    True

  • @vijaylakshminarayanan7052
    @vijaylakshminarayanan7052 3 года назад +3

    Strongly disagree I am sorry it's oversimplified! Data is important, but algorithms are the key to achieve breakthrough performance. I see a lot of these "steal the code" projects with abysmal performance in real world applications. For example you can use 1m images throw lot of GPU and train a face detection model, or improve the residual connections with a killer dlib algorithm, and implement on a small footprint like Raspberry Pi with decent frame rate!

    • @vijaylakshminarayanan7052
      @vijaylakshminarayanan7052 3 года назад +1

      @Latus gomy I don't think you understood my comment. Real life problems (for example feature learning in face detection) are rarely as simple as A+B. Its similar to a non-linear PDE. Each individual equation must be studied as a separate problem, and optimised. In my experience deploying a few computer vision projects to achieve good performance along with data, modifying the algorithms is necessary. But feel free to follow their "steal the code" approach!