Preprint announcement from David J C MacKay

[David J.C. MacKay]

The following preprints are available by anonymous ftp or www. 

WWW: The page: 
ftp://131.111.48.8/pub/mackay/README.html
has pointers to abstracts and postscript of these publications. 

-------------
 Titles
-------------

1) 	Probable Networks and Plausible Predictions - 
   	 A Review of Practical Bayesian Methods for Supervised Neural Networks 

2) 	Density Networks and their application to Protein Modelling

3) 	A Free Energy Minimization Framework for 
   	 Inference Problems in Modulo 2 Arithmetic

4) 	Interpolation models with multiple hyperparameters

--------------
 Details
--------------

1) Probable Networks and Plausible Predictions - 
   A Review of Practical Bayesian Methods for Supervised Neural Networks 

	by David J C MacKay

	Review paper to appear in `Network' (1995).

	Final version (1 Feb 95). 41 pages. (508K)
	ftp://131.111.48.8/pub/mackay/network.ps.Z

2) Density Networks and their application to Protein Modelling

	by David J C MacKay

	Abstract:

	I define a latent variable model in the form of a neural
	network for which only target outputs are specified; the
	inputs are unspecified.  Although the inputs are missing, it
	is still possible to train this model by placing a simple
	probability distribution on the unknown inputs and maximizing
	the probability of the data given the parameters.  The model
	can then discover for itself a description of the data in
	terms of an underlying latent variable space of lower
	dimensionality.  I present preliminary results of the
	application of these models to protein data.

	(to appear in Maximum Entropy 1994 Proceedings [1995])

	ftp://131.111.48.8/pub/mackay/density.ps.Z (130K)

3) A Free Energy Minimization Framework for 
   Inference Problems in Modulo 2 Arithmetic

	by David J C MacKay

	Abstract: 

	This paper studies the task of inferring a binary vector s
	given noisy observations of the binary vector t = A s mod 2,
	where A is an M times N binary matrix. This task arises in
	correlation attack on a class of stream ciphers and in other
	decoding problems.

	The unknown binary vector is replaced by a real vector of
	probabilities that are optimized by variational free energy
	minimization.  The derived algorithms converge in
	computational time of order between w_{A} and N w_{A}, where
	w_{A} is the number of 1s in the matrix A, but convergence to
	the correct solution is not guaranteed.

	Applied to error correcting codes based on sparse matrices A,
 	these algorithms give a system with empirical performance
 	comparable to that of BCH and Reed-Muller codes.

	Applied to the inference of the state of a linear feedback
	shift register given the noisy output sequence, the algorithms
	offer a principled version of Meier and Staffelbach's (1989)
	algorithm B, thereby resolving the open problem posed at the
	end of their paper. The algorithms presented here appear to
	give superior performance.

	(to appear in Proceedings of 1994 K.U. Leuven Workshop on
		  Cryptographic Algorithms)
	ftp://131.111.48.8/pub/mackay/fe.ps.Z (101K)

4) Interpolation models with multiple hyperparameters

	by David J C MacKay and Ryo Takeuchi

	Abstract: 

	A traditional interpolation model is characterized by the
 	choice of regularizer applied to the interpolant, and the
 	choice of noise model.  Typically, the regularizer has a
 	single regularization constant alpha, and the noise model
 	has a single parameter beta.  The ratio alpha/beta
 	alone is responsible for determining globally all these
 	attributes of the interpolant: its `complexity',
 	`flexibility', `smoothness', `characteristic scale length',
 	and `characteristic amplitude'.  We suggest that interpolation
 	models should be able to capture more than just one flavour of
 	simplicity and complexity.  We describe Bayesian models in
 	which the interpolant has a smoothness that varies
 	spatially. We emphasize the importance, in practical
 	implementation, of the concept of `conditional convexity' when
 	designing models with many hyperparameters.

	(submitted to IEEE PAMI)

	ftp://131.111.48.8/pub/mackay/newint.ps.Z (179K)

To get papers by anonymous ftp, follow the usual procedure: 
ftp 131.111.48.8      anonymous     cd pub/mackay     binary     get ...

------------------------------------------------------------------------------
David J.C. MacKay         email: mackay at mrao.cam.ac.uk                     
Radio Astronomy,            www: ftp://131.111.48.24/pub/mackay/homepage.html
Cavendish Laboratory,       tel: +44 1223 337238  fax: 354599    home: 276411
Madingley Road,               
Cambridge CB3 0HE. U.K.    home: 19 Thornton Road, Girton, Cambridge CB3 0NP
------------------------------------------------------------------------------