🦉🦉   WOOT WOOT!   RAPIDMINER WISDOM 2020 EARLY BIRD REGISTRATION ENDS FRIDAY DEC 13!   REGISTER NOW!   🦉🦉

موقعیت نسبی

mbsmbs Member, KB Contributor Posts: 221  Guru
سلام
وقت بخیر
برای شرایطی که نمیتوان با قاطعیت تصمیم گرفت پیشنهاد شما جز از استفاده از منطق فازی چیست؟
آیا برای شرایطی همسان با گوگل الگوریتمی بجز از یادگیر عمیق و شبکه عصبی پیشنهاد میدهید؟
متشکرم

Answers

  • sgenzersgenzer 12Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,656  Community Manager
    hello @mbs - there is no answer to this question without more specific information. What exactly are you trying to do?

    Scott
  • mbsmbs Member, KB Contributor Posts: 221  Guru
    Hello
    @sgenzer
    I have a situation like google and I want to make a software which has the same process like google and some how a bit more complex than google so in this situation what is your suggestion for an algorithm?

    Thank you 
    mbs

  • sgenzersgenzer 12Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,656  Community Manager
    sorry @mbs but that is still very vague!
  • mbsmbs Member, KB Contributor Posts: 221  Guru
    edited October 1
    @sgenzer

    Thank you that you follow the problem and for your help.

    I have a computer vision process after that I have to use machine learning on the result of that so the process is like google because it has a big database so what is the best algorithm for machine learning part?

    Many thanks

    mbs
  • rfuentealbarfuentealba Moderator, RapidMiner Certified Analyst, Member, University Professor Posts: 442   Unicorn
    Hello @mbs,

    If you have a large dataset for computer vision, there is only a handful of algorithms that you can apply. The most common are:
    • Neural Network.
    • Deep Learning.
    • Convolutional Neural Network.
    • HAAR Cascades.
    Besides that, you should define what kind of computer vision algorithms do you require: do you need edge detection? do you need depth management? do you need to transpose images so that you can "paste those" depending on the content?

    Before doing computer vision, you will need to apply some transformations depending on the depth captured by the camera or certain shapes. Are you considering those too? These processes might be expensive if not done by GPU.

    You choose your algorithms depending on what images are you using, and how you are using these. You may as well want to use channels (RGB) and check how similar are the images before deciding which ones to use.
    sgenzerTghadially
  • mbsmbs Member, KB Contributor Posts: 221  Guru

    Hello

    What do you mean by using " channels(RGB)"? plz explain more.

    Yes I need edge detection and need depth management too.  Is ensemble learning good for that? 

    I have to work on sequence of images.

    So which algorithm is better for that?

    Thanks

    mbs
    Tghadially
  • rfuentealbarfuentealba Moderator, RapidMiner Certified Analyst, Member, University Professor Posts: 442   Unicorn
    Each picture can be decomposed in red, green and blue "channels" (primary colors for painting with light, as opposed to red, yellow and blue that are primary colors for working with pigments).

    A Convolutional Neural Network is an ensemble of neural networks that acts on each channel independently. It works for object recognition, edge detection and many other things. I would advise you to use an edge detection algorithm such as Sobel, Laplacian, Prewitt or... I always forget the other one, as an input.

    For depth detection, I would advice you to search for Andrew Ng's depth detection algorithms. You will face many issues, such as the color correction (the blue and black dress that some people mistook as white and gold), but it all depends on what kinds of images are you using.

    All the best,

    Rod.
  • mbsmbs Member, KB Contributor Posts: 221  Guru
    ensemble means combining algorithms in order to make it more strong.

    For example:deep learning+nural network

    The color should be all in gry

    mbs
  • varunm1varunm1 Moderator, Member Posts: 970   Unicorn
    edited October 2
    I recommend you try 3D CNN if the images are sequential. The 3 Dimensional CNN also helps capture temporal properties (a bit) with spatial properties. You can also create fusion networks which are a combination of CNN and RNN that capture both spatial and temporal properties from data. Here is one good article related to this. 

    https://blog.coast.ai/five-video-classification-methods-implemented-in-keras-and-tensorflow-99cad29cc0b5
    rfuentealba
  • rfuentealbarfuentealba Moderator, RapidMiner Certified Analyst, Member, University Professor Posts: 442   Unicorn
    Hi @mbs,

    Yes, that is the definition of an ensemble method. To determine which ones will get you better performance, though, you should use non-ensembles first and then combine and test. There is a series of recommendations, such as not combining stable learners because these won't help you get a better prediction and a long list of things.

    All the best,

    Rod.
    varunm1
  • mbsmbs Member, KB Contributor Posts: 221  Guru
    edited October 4


    Thank you
Sign In or Register to comment.