SciTech #ScienceSunday Digest - 40/2016.
Originally shared by null
SciTech #ScienceSunday Digest - 40/2016.
Permalink here: http://www.scitechdigest.net/2016/10/machine-learning-fps-bot-antimatter.html
Machine learning FPS bot, Antimatter from lasers, Google language translation, Embryo and baby modification, Neuromorphic deep learning, Nanopore DNA sequencing, Hand exoskeleton, Autonomous gap navigation, Multiferroic materials, Imaging scattered light.
1. Machine Learning Agent Plays Doom Video Game
A new AI agent or bot built by machine learning algorithms to learn, play, and master the 3D first-person-shooter video game Doom is an expected, albeit confronting demonstration https://techcrunch.com/2016/09/21/scientists-teach-machines-to-hunt-and-kill-humans-in-doom-deathmatch-mode/. Be sure to check out the video of the bot in action; I wonder how long until real-world robots are loaded up with improved software? This seems to build on DeepMind’s approach, with a bot that interacts with the game based on what is on-screen: navigating the environment, identifying objects, and “interacting” with those objects with little to no pre-programming. The system beat both conventional in-game bots and human players.
2. Generating Antimatter from Lasers
Advances in ultrahigh intensity laser technology should soon enable the creation of electric fields so intense that matter and antimatter are generated https://publishing.aip.org/publishing/journal-highlights/creating-antimatter-lasers. Recent calculations show how to use these technologies to create electrons and positrons, in which the physicists claim to “boil the vacuum” and convert virtual electron-positron pairs into real, observable particles. Under certain conditions the light-matter interaction produces a skewed cascade that results in a large excess of positrons over electrons. A controllable, efficient source of on-demand antimatter would have a huge range of applications.
3. Human Performance for Language Translation
Google’s latest language translation tools utilise more advanced deep learning algorithms to get very close to human levels of performance https://www.technologyreview.com/s/602480/googles-new-service-translates-languages-almost-as-well-as-humans-can/. All language pairs performed 60% - 87% better than previous approaches, and in some cases getting to within just 2% of human level performance; these new tools should be rolling out to Google’s translation products soon. The software works out its own way to break up text and perform the translation and does so in ways that often don’t make sense and are not understood by the system’s creators.
4. Altering Embryo and Baby Genetics
The first baby to have DNA from three different parents has been born https://www.newscientist.com/article/2107219-exclusive-worlds-first-baby-born-with-new-3-parent-technique/. In this case the reason was to treat congenital mutations in mitochondria, and involved (i) taking the nucleus from the mother’s egg, (ii) inserting it into a donor egg whose nucleus was removed (providing healthy mitochondria), and (iii) fertilising this hybrid egg with the father’s sperm. We also had work in which (days old) healthy human embryo’s had their genes edited in order to tweak and regulate developmental processes http://www.livescience.com/56243-human-embryo-editing.html - these embryos were never intended to be allowed to develop beyond a couple of weeks.
5. IBM’s Neuromorphic Chips get Deep Learning
IBM has further developed its TrueNorth neuromorphic computing architecture to the point that it can now run deep learning algorithms http://spectrum.ieee.org/tech-talk/computing/hardware/ibms-braininspired-chip-tested-on-deep-learning. TrueNorth’s spiking neural networks are typically incompatible with deep learning algorithms but this has now been overcome with a new way of implementing the algorithms on this hardware allowing TrueNorth to match or surpass state-of-the-art accuracy on a number of tests. In related news new memristors use silver-in-oxide dynamics to mimic calcium dynamics of synapses and result in synaptic emulators for neuromorphic computing http://www.nextbigfuture.com/2016/09/memristors-as-synaptic-emulators-for.html.
6. MoS2 Nanopore DNA Sequencing
A nanopore DNA sequencer has been designed using molybdenum disulfide for the first time https://www.nist.gov/news-events/news/2016/09/nist-team-suggests-nanoscale-electronic-motion-sensor-dna-sequencer. In this design DNA nucleotide bases are connected to the inside of the nanopore, with the pore itself being a hole through a ribbon of molybdenum disulfide suspended between electrodes: as DNA is passed through the pore, complementary bases in the sequence bind and release partner bases connected to the hole, causing the ribbon to flex, the motion of which is dependent on the base binding and generates a detectable electrical signal. Estimates suggest a 79% - 85% sequence accuracy, at single pass of 70 million bases per second.
7. Muscle Sensing Hand Exoskeleton
An articulated 3D printed hand exoskeleton is worn on the back of the hand and includes electromyography sensors on the back of the forearm http://www.relab.ethz.ch/research/current-research-projects/robotic-hand-orthosis-for-therapy-and-assistance-in-activities-of-daily-living.html. The sensors detect signals in the muscles then activate motors to provide additional grasping and grip strength to mimic the intended motion of the hand. 3D printing allows customisation to different sized patients while the group intends to develop and test alternative sensors such as near-infrared spectroscopy and electroencephalography (ECG) to allow cortical or thought-controlled exoskeleton hand motion.
8. Autonomous Gap Navigation by Drones
Quadrotor drones have recently achieved completely autonomous on-board gap navigation and traversal http://spectrum.ieee.org/automaton/robotics/drones/aggressive-quadrotors-conquer-gaps-with-ultimate-autonomy. The drone uses nothing more complicated than the equivalent of a smartphone on board, and is able to locate the gap, compute an optimal trajectory, and execute the flight path to get through to the other side with currently an 80% success rate. The group hope to boost this success rate further as well as navigate any arbitrary number of consecutive gaps in a row. Ultimately this is getting closer to the goal of autonomous navigation through any arbitrarily haphazard environment of interest.
9. Multiferroic and Cold-Sintered Materials
First, layer by layer assembly of nanosheet building blocks as produced multiferroic materials that function at room temperature for the first time and which exhibit both ferroelectricity and ferromagnetism http://www.nims.go.jp/eng/news/press/2016/09/201609230.html. I haven’t come across multiferroics before and wonder what applications this phenomena enables. Second, a newly developed cold-sintering-process can densify and create an entirely new range of hybrid ceramic-like materials at relatively low temperatures that can combine materials that can’t normally be combined such as ceramic-polymers and ceramic-metals, in addition to ceramic-nanoparticles and ceramic-ceramics http://news.psu.edu/story/428704/2016/09/28/research/lowering-heat-makes-new-materials-possible-while-saving-energy.
10. Better Imaging Through Optical Obstructions
New image processing algorithms can accurately produce images of objects even after the light bouncing off of or emitted by those objects has been chaotically scattered by semi-opaque materials such as human tissue or fog before reaching the camera http://news.mit.edu/2016/all-photons-imaging-algorithm-0929. The system fires a pulsed laser at the object to be imaged, then measures the timing of different photons arriving from each pulse; those photons passing straight through reach the sensor first, while those that are scattered ever-more take ever-longer to reach the sensor and deviate from the location of the arrival of the first photons by a larger amount. Applications include medical imaging through human tissues and better navigation by autonomous vehicles through fog and rain.
SciTech Tip Jar: http://www.scitechdigest.net/p/donate.html
SciTech #ScienceSunday Digest - 40/2016.
Permalink here: http://www.scitechdigest.net/2016/10/machine-learning-fps-bot-antimatter.html
Machine learning FPS bot, Antimatter from lasers, Google language translation, Embryo and baby modification, Neuromorphic deep learning, Nanopore DNA sequencing, Hand exoskeleton, Autonomous gap navigation, Multiferroic materials, Imaging scattered light.
1. Machine Learning Agent Plays Doom Video Game
A new AI agent or bot built by machine learning algorithms to learn, play, and master the 3D first-person-shooter video game Doom is an expected, albeit confronting demonstration https://techcrunch.com/2016/09/21/scientists-teach-machines-to-hunt-and-kill-humans-in-doom-deathmatch-mode/. Be sure to check out the video of the bot in action; I wonder how long until real-world robots are loaded up with improved software? This seems to build on DeepMind’s approach, with a bot that interacts with the game based on what is on-screen: navigating the environment, identifying objects, and “interacting” with those objects with little to no pre-programming. The system beat both conventional in-game bots and human players.
2. Generating Antimatter from Lasers
Advances in ultrahigh intensity laser technology should soon enable the creation of electric fields so intense that matter and antimatter are generated https://publishing.aip.org/publishing/journal-highlights/creating-antimatter-lasers. Recent calculations show how to use these technologies to create electrons and positrons, in which the physicists claim to “boil the vacuum” and convert virtual electron-positron pairs into real, observable particles. Under certain conditions the light-matter interaction produces a skewed cascade that results in a large excess of positrons over electrons. A controllable, efficient source of on-demand antimatter would have a huge range of applications.
3. Human Performance for Language Translation
Google’s latest language translation tools utilise more advanced deep learning algorithms to get very close to human levels of performance https://www.technologyreview.com/s/602480/googles-new-service-translates-languages-almost-as-well-as-humans-can/. All language pairs performed 60% - 87% better than previous approaches, and in some cases getting to within just 2% of human level performance; these new tools should be rolling out to Google’s translation products soon. The software works out its own way to break up text and perform the translation and does so in ways that often don’t make sense and are not understood by the system’s creators.
4. Altering Embryo and Baby Genetics
The first baby to have DNA from three different parents has been born https://www.newscientist.com/article/2107219-exclusive-worlds-first-baby-born-with-new-3-parent-technique/. In this case the reason was to treat congenital mutations in mitochondria, and involved (i) taking the nucleus from the mother’s egg, (ii) inserting it into a donor egg whose nucleus was removed (providing healthy mitochondria), and (iii) fertilising this hybrid egg with the father’s sperm. We also had work in which (days old) healthy human embryo’s had their genes edited in order to tweak and regulate developmental processes http://www.livescience.com/56243-human-embryo-editing.html - these embryos were never intended to be allowed to develop beyond a couple of weeks.
5. IBM’s Neuromorphic Chips get Deep Learning
IBM has further developed its TrueNorth neuromorphic computing architecture to the point that it can now run deep learning algorithms http://spectrum.ieee.org/tech-talk/computing/hardware/ibms-braininspired-chip-tested-on-deep-learning. TrueNorth’s spiking neural networks are typically incompatible with deep learning algorithms but this has now been overcome with a new way of implementing the algorithms on this hardware allowing TrueNorth to match or surpass state-of-the-art accuracy on a number of tests. In related news new memristors use silver-in-oxide dynamics to mimic calcium dynamics of synapses and result in synaptic emulators for neuromorphic computing http://www.nextbigfuture.com/2016/09/memristors-as-synaptic-emulators-for.html.
6. MoS2 Nanopore DNA Sequencing
A nanopore DNA sequencer has been designed using molybdenum disulfide for the first time https://www.nist.gov/news-events/news/2016/09/nist-team-suggests-nanoscale-electronic-motion-sensor-dna-sequencer. In this design DNA nucleotide bases are connected to the inside of the nanopore, with the pore itself being a hole through a ribbon of molybdenum disulfide suspended between electrodes: as DNA is passed through the pore, complementary bases in the sequence bind and release partner bases connected to the hole, causing the ribbon to flex, the motion of which is dependent on the base binding and generates a detectable electrical signal. Estimates suggest a 79% - 85% sequence accuracy, at single pass of 70 million bases per second.
7. Muscle Sensing Hand Exoskeleton
An articulated 3D printed hand exoskeleton is worn on the back of the hand and includes electromyography sensors on the back of the forearm http://www.relab.ethz.ch/research/current-research-projects/robotic-hand-orthosis-for-therapy-and-assistance-in-activities-of-daily-living.html. The sensors detect signals in the muscles then activate motors to provide additional grasping and grip strength to mimic the intended motion of the hand. 3D printing allows customisation to different sized patients while the group intends to develop and test alternative sensors such as near-infrared spectroscopy and electroencephalography (ECG) to allow cortical or thought-controlled exoskeleton hand motion.
8. Autonomous Gap Navigation by Drones
Quadrotor drones have recently achieved completely autonomous on-board gap navigation and traversal http://spectrum.ieee.org/automaton/robotics/drones/aggressive-quadrotors-conquer-gaps-with-ultimate-autonomy. The drone uses nothing more complicated than the equivalent of a smartphone on board, and is able to locate the gap, compute an optimal trajectory, and execute the flight path to get through to the other side with currently an 80% success rate. The group hope to boost this success rate further as well as navigate any arbitrary number of consecutive gaps in a row. Ultimately this is getting closer to the goal of autonomous navigation through any arbitrarily haphazard environment of interest.
9. Multiferroic and Cold-Sintered Materials
First, layer by layer assembly of nanosheet building blocks as produced multiferroic materials that function at room temperature for the first time and which exhibit both ferroelectricity and ferromagnetism http://www.nims.go.jp/eng/news/press/2016/09/201609230.html. I haven’t come across multiferroics before and wonder what applications this phenomena enables. Second, a newly developed cold-sintering-process can densify and create an entirely new range of hybrid ceramic-like materials at relatively low temperatures that can combine materials that can’t normally be combined such as ceramic-polymers and ceramic-metals, in addition to ceramic-nanoparticles and ceramic-ceramics http://news.psu.edu/story/428704/2016/09/28/research/lowering-heat-makes-new-materials-possible-while-saving-energy.
10. Better Imaging Through Optical Obstructions
New image processing algorithms can accurately produce images of objects even after the light bouncing off of or emitted by those objects has been chaotically scattered by semi-opaque materials such as human tissue or fog before reaching the camera http://news.mit.edu/2016/all-photons-imaging-algorithm-0929. The system fires a pulsed laser at the object to be imaged, then measures the timing of different photons arriving from each pulse; those photons passing straight through reach the sensor first, while those that are scattered ever-more take ever-longer to reach the sensor and deviate from the location of the arrival of the first photons by a larger amount. Applications include medical imaging through human tissues and better navigation by autonomous vehicles through fog and rain.
SciTech Tip Jar: http://www.scitechdigest.net/p/donate.html
Hi Jessica Meyer MY Good Friend your research or work is awesome thank you for sharing so I have something to share with you🌹
ReplyDeleteThe Worlds First Artificial Pancreas in a few months.
Its called the MiniMed 670G,ITS A HYBRID, a automated glucose monitor and a insulin pump and the FDA approved it but further testing is needed.
Medtronic:
describes the MiniMed 670G,as the first hybrid closed loop system in the world.
The device only requires patients to input meal carbohydrates and calibrate the sensors periodically, allowing them to live life more freely.
It also helps them sleep well through the night,and wake up with healthy glucose levels.[New Science]^,^[Evalving]°_°]
🌹🌹🌹🌹🌹🌹🌹🌹🌹🌹
Hi Jessica Meyer that game looks fun I defiantly would play it thank you My Good Friend 🎮😀👍👌👊✊
ReplyDelete🌹🌹🌹🌹🌹🌹🌹🌹🌹🌹🌐