Likelihood measure

  • Nobody/Anonymous

    Is there such a thing as this? This is my aim (the field is image pattern recognition):-

    myInput -> Neural Network A -> Likelihood of being objectA
    myInput -> Neural Network B -> Likelihood of being objectB
    myInput -> Neural Network C -> Likelihood of being objectC
    myInput -> Neural Network D -> Likelihood of being objectD

    I want to train several neural networks to recognise a different particular object, run my input through each and get returned a number between 0 and 1 that says how likely the input is the object the network is trained for.

    Could someone suggest a network for this? I had a look in samples/editor/som/SOMImageNet.ser and this uses a WinnerTakesAll layer. Which obviously doesn't give me a likelihood.

    Could someone suggest an alternate approach or a way to do this?



Log in to post a comment.