Dear Hoover,
I posted a query on CVNet about visual model based feedback neural networks
on June 10. To my pleasure, some colleagues sent their helpful ideas to me.
However, it seems that not many models in this area have been proposed, though
they can be quite interesting and inspiring. There have been a lot of feedback
phenomena which were reported in bio-vision systems. Deeper understanding in
them can be got from careful consideration on the above topic.
Some people asked me for the search results. The following is the editted
messages I received. I wish it can be useful for somebody interested in this
field. Would you please forward it on CVNet for me? Thanks.
I would like to express my thanks to the participators of this discussion for
their comments and papers.
Regards,
Hanchuan Peng
-- ------------------------------------------------------------------------- Hanchuan Peng Ph.D. Student | Tel: 86-25-3619983 Department of Biomedical Engineering | Fax: 86-25-7712719 Southeast University | Email: phc@seu.edu.cn Nanjing, 210096, P.R.China | URL: http://lmbe.seu.edu.cn/phc/ -------------------------------------------------------------------------------------------------------------------------------------------------- >From francis@RUCKER-EMH2.ARMY.MIL Mon Jun 10 23:21:57 1996Subject: feedback neural networks
Hi,
I saw your posting on CVNET this morning asking for info on feedback neural network models of visual perception. I am out of my office this summer, so I can only give you as much of a reference as I remember.
The most developed neural network model of visual perception is the Boundary Contour System (BCS) created by Stephen Grossberg and his colleagues at Boston University. This sytem uses feedback to complete contour (boundary) information across noise, retinal veins,... Look in Grossberg (1994), _Perception & Psychophysics_ for the latest developments.
I have worked with this model (I am one of Grossberg's students) to study its dynamic characteristics. The feedback is critical to my findings. The feedback supports lasting persistence of neural responses. Other mechanisms act to cut short that persistence to prevent image smearing. The resulting interactions account for a lot of psychophysical data. Look in Francis, Grossberg & Mingolla (1994) _Vision Research_, Francis & Grossberg (1996) _Vision Research_.
You can also find some in-press articles on my web page at http://www.psych.purdue.edu/~gfrancis/home.html
Hope this helps.
-Greg Francis Assistant Professor, Psychological Sciences Purdue University gfrancis@psych.purdue.edu
----------------------------------------------------------------------- >From hrw6@midway.uchicago.edu Tue Jun 11 06:30:34 1996Subject: Nonlinear Feedback and Vision
Dear Mr. Peng: I have published work on nonlinear feedback and the psychophysics of contrast gain control circuits: Wilson, H. R. and Humanski, R. (1993) Spatial frequency adaptation and contrast gain control. Vision Res. 33, 1133-1149.
I have also developed a detailed, quantitative neural model for two-dimensional motion perception that incorporates negative feedback to decide whether the stimnulus motion is coherent or transparent: Wilson, H. R. and Kim, J. (1994) Perceived motion in the vector sum direction. Vision Res. 34, 1835-1842. Wilson, H. R. and Kim, J. (1994) A model for motion coherence and transparency. Visual Neurosci. 11, 1205-1220. Wilson, H. R. (1994) Models of two-dimensional motion perception. In Visual Detection of Motion, ed. by A. T. Smith & R. J. Snowden, Academic Press, London, 219-251.
Finally, in 1972 and 1973 Wilson (i.e. myself) and Cowan published a general model of both positive and negative feedback in cortical tissue. The reference is in Biological Cybernetics, and it is referenced in some of the above works. Best wishes, Hugh R. Wilson
>From rao@cs.rochester.edu Tue Jun 11 05:58:01 1996Subject: vision models and neural networks
Hi, This is with reference to your inquiry regarding vision models and the role of feedback. You will find some relevant references in our recent papers that can obtained via ftp as given below. Let me know if you have any problems.
-- Rajesh Rao Internet: rao@cs.rochester.edu Dept. of Computer Science VOX: (716) 275-2527 University of Rochester FAX: (716) 461-2018 Rochester NY 14627-0226 WWW: http://www.cs.rochester.edu/u/rao/==========================================================================
Dynamic Model of Visual Recognition Predicts Neural Response Properties In The Visual Cortex
Rajesh P.N. Rao and Dana H. Ballard
Technical Report 95.4 (revised) National Resource Laboratory for the study of Brain and Behavior University of Rochester, Rochester, NY 14627
(Also, Neural Computation - in review)
Abstract
The responses of visual cortical neurons during fixation tasks can be significantly modulated by stimuli from beyond the classical receptive field. Modulatory effects in neural responses have also been recently reported in a task where a monkey freely views a natural scene. In this paper, we describe a stochastic network model of visual recognition that explains these experimental observations by using a hierarchical form of the extended Kalman filter as given by the Minimum Description Length (MDL) principle. The model dynamically combines input-driven bottom-up signals with expectation-driven top-down signals to predict current recognition state. Synaptic weights in the model are adapted in a Hebbian manner according to a stochastic learning rule also derived from the MDL principle. The architecture of the model posits an active computational role for the reciprocal connections between adjoining visual cortical areas in determining neural response properties. In particular, the model demonstrates the possible role of feedback from higher cortical areas in mediating neurophysiological effects due to stimuli from beyond the classical receptive field. Simulations of the model are provided that help explain the experimental observations regarding neural responses in both free viewing and fixating conditions.
Retrieval information:
FTP-host: ftp.cs.rochester.edu FTP-pathname: /pub/u/rao/papers/dynmem.ps.Z WWW URL: ftp://ftp.cs.rochester.edu/pub/u/rao/papers/dynmem.ps.Z
32 pages; 534K compressed.
==========================================================================
A Class of Stochastic Models for Invariant Recognition, Motion, and Stereo
Rajesh P.N. Rao and Dana H. Ballard
Technical Report 96.5 National Resource Laboratory for the study of Brain and Behavior University of Rochester, Rochester, NY 14627
(Submitted to NIPS 1996)
Abstract
We describe a general framework for modeling transformations in the image plane using a stochastic generative model. Algorithms that resemble the well-known Kalman filter are derived from the MDL principle for estimating both the generative weights and the current transformation state. The generative model is assumed to be implemented in cortical feedback pathways while the feedforward pathways implement an approximate inverse model to facilitate the estimation of current state. Using the above framework, we derive stochastic models for invariant recognition, motion estimation, and stereopsis, and present preliminary simulation results demonstrating recognition of objects in the presence of translations, rotations and scale changes.
Retrieval information:
FTP-host: ftp.cs.rochester.edu FTP-pathname: /pub/u/rao/papers/invar.ps.Z URL: ftp://ftp.cs.rochester.edu/pub/u/rao/papers/invar.ps.Z
7 pages; 430K compressed.
==========================================================================
Detailed anonymous ftp instructions:
>ftp ftp.cs.rochester.eduConnected to anon.cs.rochester.edu. 220 anon.cs.rochester.edu FTP server (Version wu-2.4(3)) ready.
Name: [type 'anonymous' here] 331 Guest login ok, send your complete e-mail address as password.
Password: [type your e-mail address here]
ftp> cd /pub/u/rao/papers/ ftp> get dynmem.ps ftp> get invar.ps ftp> bye
---------------------------------------------------------------------------- >From slehar@cns.bu.edu Tue Jun 11 04:00:31 1996Subject: Feedback in Vision Models
Hi Han-chuan,
I am writing in response to your posting on CVNet on feedback in vision models. I have been interested in that same subject for many years now.
Zucker [1] has done some work on relaxation algorithms for enhancing lines and curves, but my greatest inspiration has been Grossberg [2], who has explicitly tried to express Gestalt principles in a neural network model.
In my own work I addressed the problem of feedback in vision models directly with a multi-layer feedforward/feedback model [3,4], and in my thesis work [5] I used a feedback neural network model to model the perception of a number of visual illusions.
In my latest paper [6] I address the problem of feedback in vision from a global perspective, discussing how different visual modalities must be coupled into a single dynamic relaxation system.
You can get copies of some of my stuff by anonymous ftp from...
ftp cns-ftp.bu.edu cd pub/slehar get README
where the README file describes the files that are available by anonymous ftp in compressed PostScript format. You might also wish to peruse my web site...
http://cns-ftp/pub/slehar/Lehar.html
in which I have (under Plato's Cave) an outline of a book I am curently working on, in a hypertext document, which is directly related to the issue of feedback in vision.
Please let me know if I can be of further assistance,
Steve Lehar
REFERENCES
[1] S. W. Zucker, R. A. Hummel, A. Rosenfeld (1977) An Application of Relaxation Labelling to Line and Curve Enhancement IEEE Trans. Comput. C-26, 393-403, 922-929
[2] Grossberg, Stephen and Ennio Mingolla, "Neural Dynamics of Per- ceptual Grouping: Textures, Boundaries, and Emergent Segmen- tations," Perception & Psychophysics, vol. 38, no. 2, pp. 141-171, 1985. [3] Lehar S., Worth A. MULTIPLE RESONANT BOUNDARY CONTOUR SYSTEM. Proceedings of the SPIE, Vol. 1469 (Applications of Artificial Neural Networks Conference) April 1991.
[4] Lehar S., Worth A. (1991) MULTI RESONANT BOUNDARY CONTOUR SYSTEM, Boston University, Center for Adaptive Systems technical report # CAS/CNS-TR-91-017
[5] Lehar S. DIRECTED DIFFUSION AND ORIENTATIONAL HARMONICS: LONG RANGE BOUNDARY COMPLETION THROUGH SHORT RANGE INTERACTIONS. PhD Thesis, Boston University, Jan 1994. UNOFFICIAL copy available from my ftp directory.
[6] Lehar S. (submitted to Perception) "A Gestalt Bubble Model of the Interaction of Lightness, Brightness, and Form Perception" preprint available from my ftp directory.
---------------------------------------------------------------------------- >From rpinter@maxwell.ee.washington.edu Wed Jun 12 03:19:41 1996Subject: feedback, vision models.
Dear H-c Peng, As a beginning, you can try my two papers (but there are many later ones too): Pinter, R.B. (1984) Adaptation of receptifve field spatial organization by multiplicative lateral inhibition. Journal of Theoretical Biology, 105, 233-243,. Pinter, R.B. (1985) Adaptation of spatial modulation transfere functions via nonlinear lateral inhibition. Biological Cybertnetics, 51, 285-291. If it isimpossible to find these, I will send copies. Regards, Prof. Pinter
------------------------------------------------------------------------------