(no subject)
Jun. 14th, 2005 12:39 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
anhinga_anhingaContinuing to write on BU neuroconference...
The very first tutorial was by Bruno Olshausen. (Bruno is a close collaborator of Jeff Hawkins, the author of the memory-prediction framework theory described in his book "On Intelligence".)
The title was "Natural image statistics and efficient neural representation", and the most relevant paper in connection with this tutorial is his paper with David Field, "Sparse coding of sensory inputs".
But I found another paper by the same authors to be especially enlightening: "How close are we to understanding V1?" If you can only read a few pages in neuroscience this year, read section 2.1 of this paper (pages 4-7), to get a picture of the rather sorry current state of mainstream neuroscience. Basically, it is commonly thought that we understand a lot about the primary visual cortical area V1, but the authors make a good argument to the contrary. The previous version of this paper was called "What is the other 85% of V1 doing?"
The very first tutorial was by Bruno Olshausen. (Bruno is a close collaborator of Jeff Hawkins, the author of the memory-prediction framework theory described in his book "On Intelligence".)
The title was "Natural image statistics and efficient neural representation", and the most relevant paper in connection with this tutorial is his paper with David Field, "Sparse coding of sensory inputs".
But I found another paper by the same authors to be especially enlightening: "How close are we to understanding V1?" If you can only read a few pages in neuroscience this year, read section 2.1 of this paper (pages 4-7), to get a picture of the rather sorry current state of mainstream neuroscience. Basically, it is commonly thought that we understand a lot about the primary visual cortical area V1, but the authors make a good argument to the contrary. The previous version of this paper was called "What is the other 85% of V1 doing?"
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
What is the other 85% of V1 doing?
Date: 2005-06-14 06:40 pm (UTC)Re: What is the other 85% of V1 doing?
Date: 2005-06-14 07:23 pm (UTC)I suppose that now when people are paying attention already, the title can be made less sensational...
Re: What is the other 85% of V1 doing?
Date: 2005-06-14 07:27 pm (UTC)no subject
Date: 2005-07-08 01:40 pm (UTC)another paper based on Olshausen&Field approach.
no subject
Date: 2005-07-09 03:29 am (UTC)Do you know what was the value of "lambda", and what is the informal explation of this particular dependency on "mu"?
no subject
Date: 2005-07-11 08:07 am (UTC)Sorry, я на работе бываю редко, где интернет.
См. там: http://redwood.ucdavis.edu/bruno/papers/
Olshausen BA, Field DJ (1997). Vision Research, 37: 3311-3325.
Там в формуле (14) лямбда - здесь мю.
Там в формуле (18) ренормализация на каждом шаге
(объясняется почему) - здесь регуляризация (.5<лямбда<1.5,
цифры мю и лямбда для базисных функций нормированных на 1).
Смысл мю (sparsenes) объясняется там же и в статьях Field D.J.
впервые введен Barlow H.B. (1961).
Ключевые имена работавших над этим (+..):
S.P.Luttrel, E.Oja, A.Hyvaerinen,
S.Amari, A.Bell, T.Sejnowski,
R.Linsker, K.Obermayer (V1 modeling)
Смысл данной статьи, в самоорганизации под влиянием
потока естественных картинок. Однако R.Linsker предложил
sparse coding model самоорганизации receptive fields and maps
у слепых котят, noise-driven (1987-89):
R.Linsker "Self-organization in a perceptual network" Computer, 21:105-117
no subject
Date: 2005-07-11 03:54 pm (UTC)