SciTech #ScienceSunday Digest - 48/2016.

SciTech #ScienceSunday Digest - 48/2016.
Permalink here: http://www.scitechdigest.net/2016/11/deep-learning-criminal-recognition.html

Deep learning criminal recognition, Mapping the epigenome, Friction on graphene, Single atom memory, Parabiosis twist, Thermoelectric paint, MEMS ultrasound interfaces, Reducing water from air, Nanochannel genome mapping, Inducing mitophagy.

1. Criminal Face Recognition by Deep Learning
A deep learning system, after being suitably trained on 1,500 images of faces, half of which were criminals, half of which were not, can predict which subsequent faces are criminals with 90% accuracy https://www.technologyreview.com/s/602955/neural-network-learns-to-identify-criminals-by-their-faces/. This follows work in 2011 that showed that humans are also adept at picking criminals from a random collection of faces. I’ll emphasise the fact that there are no guarantees in biology, just predispositions and likelihoods, and that a lot more work is needed, but this opens up some pretty interesting (potentially worrying) possibilities in future with regards to automatic screening for measures of criminality, trustworthiness, and other factors, either by individuals or by states. Imagine a Facebook plug-in that tagged everyone based on these assessments, or surveillance systems that directed personnel attention to specific individuals.

2. Mapping the Human Epigenome
The human epigenome, that pattern of methylation tags on DNA that help regulate gene expression, has been mapped to extraordinary detail in BLUEPRINT studies on how hematopoietic stem cells differentiate and produce the various types of blood cells https://www.eurekalert.org/pub_releases/2016-11/crcf-btd111716.php. Different epigenomic maps help dictate the particular gene expression blueprint that is to be followed to produce cells of a particular type, while disruption or changes to epigenetic markers can often result in disease states. Apart from raw knowledge, applications include cancer diagnostics, personalised medicine, and improved cellular reprogramming.

3. Understanding Graphene Friction
The possible applications for using graphene and graphite as lubricants takes a step forward with a far more sophisticated understanding of how friction operates on graphene surfaces http://news.mit.edu/2016/sliding-flexible-graphene-surfaces-1123. It turns out that in addition to quantity of contact (how much area between the two surfaces actually touch) the group discovered that quality of contact (how well individual carbon atoms make contact with other atoms in the material) plays a large role and causing the counter-intuitive spike in friction as movement begins before levelling off. The new understanding opens up possibilities in tuning graphene interfaces to provide a desired level of friction. In other 2D materials news, indium selenide appears to offer some interesting “goldilocks” properties http://spectrum.ieee.org/nanoclast/semiconductors/materials/indium-selenide-takes-on-the-mantle-of-the-new-wonder-material.

4. Single Atom Magnet Superlattice
Building on single-atom work announced earlier this year a group has now produced monomer layer superlattices of single (dysprosium) atom magnets on graphene-iridium sheets with a theoretical information density of 115 terabits per square inch http://phys.org/news/2016-11-superlattice-single-atom-magnets-aims-ultimate.html. The surfaces are prepared at 40K, allowing the atoms to find the lowest energy states in the lattice with an atomic spacing of 2.5nm, although stable magnet performance is only possible below 10K. This very low temperature prototype isn’t going to be useful anytime soon, which will require massive improvements in temperature stability.

5. Another Parabiosis Twist
Following on from the interesting parabiosis work last week showing rejuvenation effects of young blood on old mice, this latest parabiosis work with a counter-narrative deserves a strong mention. This latest study on mice, conducted with a different blood transfer procedure, suggests that young blood does not by itself help rejuvenate older animals and that older blood in young mice causes declines in the functioning of most organs and tissues http://news.berkeley.edu/2016/11/22/young-blood-does-not-reverse-aging-in-old-mice-uc-berkeley-study-finds/. This suggests that blood transfusions using blood from older individuals may carry a number of risks. Also, it appears that better ways to filter blood or otherwise remove accumulating factors in older blood will lead to rejuvenation benefits for older individuals.

6. Thermoelectric Paint
Thermoelectric paints (that convert heat into electricity) that can be quickly applied to any surface have been developed for the first time, in order to generate electricity from large temperature differences http://news.unist.ac.kr/unist-engineers-thermoelectric-material-in-paintable-liquid-form/. Like paint-on photovoltaics and LEDs this is a good initial demonstration that provides a platform for development: better efficiencies, lower operating temperatures, etc. Applications include any machine subject to high temperatures including cars, satellites, computer chips, and many others.

7. MEMS Ultrasound Interfaces
Recent advances and improvements in MEMS technology and devices will soon result in consumer level MEMS chips that function as ultrasonic gesture interfaces http://spectrum.ieee.org/semiconductors/devices/beyond-touch-tomorrows-devices-will-use-mems-ultrasound-to-hear-your-gestures. This is like Google’s Soli technology (radar) or infrared LEDs, but makes use of ultrasound to do the same and much more at much lower power and avoiding a range of potential interference sources from the environment. Earlier MEMS-based ultrasound devices include digital pens and styluses, but the new devices coming incorporate embedded piezoelectrics and promise a whole new way to interact with our digital devices and sensors.

8. Easy Access to Water from Air
A great little student project developed a simple device, predominantly 3D printed, that is able to produce just under two litres of water per hour from humid air (lower volumes from drier air) http://newworldwow.com/index.php/2016/11/17/student-creates-3d-printed-gadget-turns-air-1-8-liters-drinkable-water-just-one-hour/. The electric fans incorporated into the device circulate air and cool it below its dew point, producing water vapour that is subsequently collected, and all while powered by a 12 volt energy source. As far as compact atmospheric water generators go this is pretty neat. Perhaps Tesla should make a similar device to be sold alongside its Power Wall energy storage batteries, a Water Wall for on-site water production from excess energy.

9. Genome Mapping with Nanochannels
While nanopores are showing great promise as a long-read DNA sequencing technology, a new device and approach using nanochannels is showing great promise for high-resolution genome mapping as a useful complementary technology http://www.nanowerk.com/spotlight/spotid=45129.php. In this approach photostable fluorophores are combined with blinking dyes to mitigate the effects of thermal fluctuations of DNA to facilitate a 15-fold improvement in mapping resolution. The benefits of rapid genome mapping include determining gene location and distance, genetic rearrangements, better accuracy for discriminating repeat numbers of long strings of identical bases that sequencing struggles to resolve.

10. Clearing Defective Mitochondria
Increasing numbers of defective mitochondria in our cells are one of the major causes of aging-related damage; while work continues to try migrate mitochondrial genes to the nucleus to mitigate this, interim measures are also being developed including co-opting mitophagy, the cells natural quality control mechanism, to induce cells to remove defective mitochondria https://www.fightaging.org/archives/2016/11/manipulating-existing-methods-of-cellular-quality-control-to-clear-mutant-mitochondria/. Recent work demonstrates several different interventions that might be developed into human therapies in future in order to accelerate and boost this quality control mechanism.

SciTech Tip Jar: http://www.scitechdigest.net/p/donate.html

Comments

  1. With regards to the neural net that they claim can identify criminals: assuming that the people in the criminal set were all convicted criminals (the paper doesn't make it clear that that is the case), I'm inclined to say that this network predicts only whether a person is likely to be convicted of a crime, not whether they are likely to commit one. IMO it's likely that it's picking up on the biases exhibited by judges in deciding whether or not to convict.

    Either way, I'd like to see it replicated.

    ReplyDelete

Post a Comment

Popular posts from this blog

#vegetarian #vegan #evolution