Efron on Fisher

A very interesting and thought-provoking article by Bradley Efron on the place of R.A. Fisher’s philosophies in the statistics of today, characterized by computational brute force:

 

http://www.jstor.org/pss/2676745

 

 

Regularization in Vision

In the 1985 Nature review article “Computational vision and regularization theory” (http://www.nature.com/nature/journal/v317/n6035/pdf/317314a0.pdf) by Tomaso Poggio, Vincent Torre and Christof Koch, which outlines the role of regularization methods in finding plausible solutions to computer vision problems, the authors have this very interesting speculation about biological vision:

 

“One of the mysteries of biological vision is its speed. Parallel processing has often been advocated as the answer to this problem. The model of computation provided by digital processes is, however, unsatisfactory, especially given the increasing evidence that neurones (sic) are complex devices, very different from simple digital switches. It is, therefore, interesting to consider whether the regularization approach to early vision may lead to a different type of parallel computation. We have recently suggested that linear, analog networks (either electrical or chemical) are, in fact, a natural way of solving the variational principles dictated by standard regularization theory.”

 

Not only do the authors provide a plausible reason for the computational speed for biological vision, but they also provide a compelling argument against the AI philosophy of treating intelligence and cognitive processes as separate from their biological/physical substrates.

 

 

 

Dirty Pictures

A very fundamental work on understanding and analyzing images under a strict
statistical framework, specifically an interpretation of standard image
processing problems under Markov field theory and Bayesian solutions:

On the Statistical Analysis of Dirty Pictures
by Julian Besag
Journal of the Royal Statistical Society. Series B (Methodological), Vol. 48,
No. 3 (1986), pp. 259-302

http://www.jstor.org/stable/pdfplus/2345426.pdf

ICML Discussion Page

I really like this idea. Each page has details of the paper and a discussion thread.

http://www.conflate.net/icml/

It’s a good way to get feedback and also increases accountability on the part of the authors

Shantanu

(Super) Fast functional imaging

An interesting paper in this month’s NeuroImage on a technique called Inverse Imaging for BOLD fMRI that reports 100ms volume acquisition times from multiple-coil arrays.

 

Lin, Fa-Hsuan; Witzel, Thomas; Mandeville, Joseph B; Polimeni, Jonathan R.; Zeffiro, Thomas A.; Greve, Douglas N.; Wiggins, Graham; Wald, Lawrence L.; Belliveau, John W. “Event-related single-shot volumetric functional magnetic resonance inverse imaging of visual processing” Neuroimage, Volume 42, issue 1 (August 1, 2008), p. 230-247

 

Abstract:

 

Developments in multi-channel radio-frequency (RF) coil array technology have enabled functional magnetic resonance imaging (fMRI) with higher degrees of spatial and temporal resolution. While modest improvement in temporal acceleration has been achieved by increasing the number of RF coils, the maximum attainable acceleration in parallel MRI acqisition is intrinsically limited only by the amount of independent spatial information in the combined array channels. Since the geometric configuration of a large-n MRI head coil array is similar to that used in EEG electrode or MEG SQUID sensor arrays, the source localization algorithms used in MEG or EEG source imaging can be extended to also process MRI coil array data, resulting in greatly improved temporal resolution by minimizing k-space traversal during signal acquisition. Using a novel approach, we acquire multi-channel MRI head coil array data and then apply inverse reconstruction methods to obtain volumetric fMRI estimates of blood oxygenation level dependent (BOLD) contrast at unprecedented whole-brain acquisition rates of 100 ms. We call this combination of techniques magnetic resonance Inverse Imaging (InI), a method that provides estimates of dynamic spatially-resolved signal change that can be used to construct statistical maps of task-related brain activity. We demonstrate the sensitivity and inter-subject reliability of volumetric InI using an event-related design to probe the hemodynamic signal modulations in primary visual cortex. Robust results from both single subject and group analyses demonstrate the sensitivity and feasibility of using volumetric InI in high temporal resolution investigations of human brain function.