New JonesBlog update. Bionic implants

The device seen in these images is called the Utah Electrode Array (WARNING: potentially graphic image after the jump of an implant in a human brain). The Utah Electrode Array is a brain implant technology developed here at the University of Utah by Richard Normann. The purpose of this device, built by currently built for us by Blackrock Microsystems is to transduce signals from external devices to deliver to the brain for interpretation. Alternatively, the device can record impulses generated in the brain for delivery of neural signals to external devices. Our potential interests in this approach are manifold, but real use and implementation of these devices is some years away still.

New JonesBlog update. Bonneville Speed Week 2009

New JonesBlog update. Its spelt F-L-Y

New JonesBlog update. USS Toledo SSN-769

The venerable BBC is reporting that a survey of light emitted from white dwarfs showed that between 1% and 3% had material (such as silicon) falling into the star on a continuous basis, potential evidence of dead worlds and asteroids. On this basis, the authors of the study speculate that the same percentage of mainstream stars in the active part of their life will have rocky matter. This is not firm evidence of actual planetary formation, as asteroids would produce the same results, but it does give an upper bound and some idea of what a lower bound might be for planetary formation.

Aside from being a useful value for Drake's Equation, the rate of planetary formation would be valuable in understanding how solar systems develop and what sort of preconditions are required for an accretion disk of suitable material to form.

Because the test only looked for elements too heavy to have been formed in the star, we can rule out the observations being that of cometary debris.

New JonesBlog update. A Computational Framework for Ultrastructural Mapping of Neural Circuitry

We have just published a manuscript in PLoS Biology where we describe how to build a complete and accurate neural network. This of course is one of the long standing holy grails in neuroscience. So, this effort meets two goals: 1) It meets the goals of building a complete neural connectome (we'll be finished collecting all of the data with cell identity, physiologic response and all synaptic connectivity in approximately six days) and 2) It defines a workflow whereby investigators from around the planet can download and use the tools we are providing to build their own connectome projects using existing infrastructure. We are making those tools available here to enable other groups to assemble, browse and annotate the terabyte sized datasets required of connectome level projects.

Four fireballs, glowing blue and orange, were visible last night over the skies of the Carolinas on the southeast coast of the United States, followed by the sound of an explosion described as being like thunder. Reports of hearing the noise were coming in from as far afield as Connecticut. There is currently no word from NASA or the USAF as to what it could be, but it seems improbable that anything non-nuclear the military could put up could be heard over that kind of distance. It therefore seems likely to be a very big meteorite.

The next question would be what type of meteorite. This is not an idle question. The one slamming into the Sudan recently was (a) extremely big at an estimated 80 tonnes, and (b) from the extremely rare F-class of asteroid. If this new meteorite is also from an F-class asteroid, then it is likely associated with the one that hit Sudan. This is important as it means we might want to be looking very closely for other fragments yet to hit.

The colours are interesting and allow us to limit what the composition could have been and therefore where it came from. We can deduce this because anything slamming through the atmosphere is basically undergoing a giant version of your basic chemistry "flame test" for substance identification. We simply need to look up what metals produce blue, and in so doing we see that cadmium does produce a blue/violet colour, with copper producing more of a blue/green.

Other metals also produce a blue glow and tables of these colours abound, but some are more likely in meteoric material than others. Cadmium exists in meteorites. Well, all elements do, if you find enough meteorites. but it exists in sufficient quantity that it could produce this sort of effect. (As noted in the chemmaster link, low concentrations can't be detected by this method, however this is going to be vastly worsened by the fact that this isn't a bunsen burner being used and the distance over which you're observing is extreme.)

Ok, what else do we know? The fireballs were also orange. Urelites, such as the Sudan impact, contain a great deal of calcium, which burns brick-red, not orange. This suggests we can rule out the same source, which in turn means we probably don't have to worry about being strafed the way Jupiter was with the Shoemaker-Levy comet (21 impacts).

What can we say about it, though? Well, provided the surviving fragments didn't fall into the ocean, it means every meteorite hunter on the planet will be scouring newspaper stories that might indicate where impacts occurred. Meteoric material is valuable and anything on a scale big enough to be heard across the entire east coast of the US is going to be worth looking for. It had split into four in the upper atmosphere, so you're probably looking at a few thousand fragments reaching ground level that would exceed a year's average pay.

New JonesBlog update. Desire...

New JonesBlog update. Urban Deer

New JonesBlog update. Photowalking Utah Hill AFB Museum

New JonesBlog update. Happy Holidays 2008

New JonesBlog update. Full Moon

New JonesBlog entry on a visit to the Korean DMZ here.

This is a glimpse at an environment that hopefully will be rapidly changing, but with North Korea going through a spasm of Communist retrenching and the uncertainty of Kim Jong Il's health (or even if he is still alive), things in the DMZ appear to be just as tense as they have been for years.

New JonesBlog entries on China here.

here

here

here

here

and here

In light of the NIST complaint that there are so many applicants for their cryptographic hash challenge that a good evaluation cannot be given, I am curious as to whether they have adequately defined the challenge in the first place. If the criteria are too loose, then of course they will get entries that are unsuitable. However, the number of hashes entered do not seem to be significantly more than the number of encryption modes entered in the encryption mode challenge. If this is impossible for them to evaluate well, then maybe that was also, in which case maybe we should take their recommendations over encryption modes with a pinch of salt. If, however, they are confident in the security and performance of their encryption mode selections, what is their real objection in the hashing challenge case?

But another question one must ask is why there are so many applicants for this, when NESSIE (the European version of this challenge) managed just one? Has the mathematics become suddenly easier? Was this challenge better-promoted? (In which case, why did Slashdot only mention it on the day it closed?) Were the Europeans' criteria that much tougher to meet? If so, why did NIST loosen the requirements so much that they were overwhelmed?

These questions, and others, look doomed to not be seriously answered. However, we can take a stab at the criteria and evaluation problem. A strong cryptographic hash must have certain mathematical properties. For example, the distance between any two distinct inputs must be unconnected to the distance between the corresponding outputs. Otherwise, knowing the output for a known input and the output for an unknown input will tell you something about the unknown input, which you don't want. If you have a large enough number of inputs and plot the distance of inputs in relation to the distance in outputs, you should get a completely random scatter-plot. Also, if you take a large enough number of inputs at fixed intervals, the distance between the corresponding outputs should be a uniform distribution. Since you can't reasonably test 2^512 inputs, you can only apply statistical tests on a reasonable subset and see if the probability that you have the expected patterns is within your desired limits. These two tests can be done automatically. Any hash that exhibits a skew that could expose information can then be rejected equally automatically.

This is a trivial example. There will be other tests that can also be applied automatically that can weed out the more obviously flawed hashing algorithms. But this raises an important question. If you can filter out the more problematic entries automatically, why does NIST have a problem with the number of entries per-se? They might legitimately have a problem with the number of GOOD entries, but even then all they need to do is have multiple levels of acceptance and an additional round or two. eg: At the end of human analysis round 2, NIST might qualify all hashes that are successful at that level as "sensitive-grade" with respect to FIPS compliance, so that people can actually start using them, then have a round 3 which produces a pool of 3-4 hashes that are "classified-grade" and a final round to produce the "definitive SHA-3". By adding more rounds, it takes longer, but by producing lower-grade certifications, the extra time needed to perform a thorough cryptanalysis isn't going to impede those who actually use such functions.

(Yes, it means vendors will need to support more functions. Cry me a river. At the current scale of ICs, you can put one hell of a lot of hash functions onto one chip, and have one hell of a lot of instances of each. Software implementations are just as flexible, with many libraries supporting a huge range. Yes, validating will be more expensive, but it won't take any longer if the implementations are orthogonal, as they won't interact. If you can prove that, then one function or a hundred will take about the same time to validate to accepted standards. If the implementations are correctly designed and documented, then proving the design against the theory and then the implementation against the design should be relatively cheap. It's crappy programming styles that make validation expensive, and if you make crappy programming too expensive for commercial vendors, I can't see there being any problems for anyone other than cheap-minded PHBs - and they deserve to have problems.)