Algorithms: The New Props for Humankind

Did  our forefathers, the hunters and gatherers,  hear better and smell better? Probably yes, alludes  Yuval Harari.  Because their survival depended on being aware of strange, hidden, prowling  dangers. Later, as hunters settled down to be farmers, life styles changed, acute sense of smell and hearing was  no longer needed and hence  eventually lost.  Yuval Harari in his  voluminous books The Sapiens and the Homo Deus, cautions us that  we stand to  lose  our power of cognition, perception and  intuition because  we are steadily outsourcing  these mental exercises  to machines and trusting their decisions implicitly. 

True, but only when dealing with huge volumes of data.  Algorithmic analyses  with their powerful pattern finding and predictive ability have become  great enablers in diverse research  fields.  A computer algorithm can in a jiffy sift through  millions of details and arrive at  the most objective decision.  Elizabeth Honig, Art Historian  at the University of California, Berkeley ropes in  the help of algorithms   to identify the  original from fake among  art pieces. Specialists in Computer Vision and Machine Learning  consider this  a win-win situation because such  exercises challenge and sharpen  an algorithm's pattern matching ability thus  making AI  an enormously powerful tool.  

Tomasev et al   too followed the same path  to  develop  a clinical tool to predict impending kidney failure. Deep Learning is an algorithm commonly used  to identify patterns in huge data sets. Tomasev and team collected well over 6 billion data points  from multiple sources from   700,000 anonymised  patients spread over 5 years - 2011 to 2015.  Special programs such as   recurrent neural network and ablation analysis were incorporated in the data analysis. Tomasev and colleagues could  register a prediction  accuracy of 84% for serious damage 90% for eventual dialysis treatment . However this retrospective analysis has to be upgraded to prospective mode.  

But there has to be checks and balances cautions, Patrick Riley Principal Engineer and Senior researcher at the Google Accelerated Science team at Google. "Many algorithms are so complicated that it is impossible to inspect all the parameters or reasons about how the inputs have been manipulated.  As these  algorithms begin    to be applied ever more widely, risks of misinterpretations, erroneous conclusions and wasted scientific efforts will spiral" he says
 

Alan Turing 1951 Courtesy: Wikipedia 

TAILPIECE:
Father of modern computing science, Alan Turing will be seen in the new £50 notes to be issued by Bank of England.   Much maligned during his life time for being gay, he  died (or committed suicide, some believe) in 1954.  Six decades later in 2013,  he was exonerated  off all blemishes by the British Queen. 

References:

1. Sapiens : A Brief History of Humankind-  Yuval Noah Harari , Harvill Secker, London 2014

2. Homo Deus :A Brief History of Tomorrow-   Yuval Noah Harari, Vintage Publications 2017 
3. Art Attribution: AI enters the fra-y  David Adams,  Nature: 13 June 2019, vol.570, pp161-162

4. A clinically applicable approach to continuous prediction of future acute kidney failure :  
    Tomasev et al Nature  1 Aug. 2019, Vol.572, pp116-119
5. Three pitfalls to avoid in machine learning.: Patrick Riley, Nature, 1 Aug.2019, vol.572,  
   pp27-28.
6. Alan Turing to be the face of new £50 note

Eavesdropping 4G!

Human Activity Recognition (HAR, for short) is an  important   component of  surveillance and security programs. HAR can be  easily, directly  and effectively done through surveillance  cameras whether  indoors or outdoors.   However cameras are helpless when the space/room  to be monitored  is  dark, dim lit, hazy or  smoke filled.  What is the alternative?  That is the question  Guo et al asked themselves and sure enough  they hit upon an answer. 

They suggest using ultrasonic sensors. Their recent paper in Applied Physics Letters describes how this can be done.  Sound waves can  propagate  unhindered even if the ambience is  dark, dim lit, hazy or  smoke filled.  Guo's team  used  a two dimensional array of  acoustic receivers and  a special algorithm based on  the CNN (Convolutional Neural Network) to process the signals. CNNs have been in use for gesture recognition and body movements of humans  engaged in routine activities. CNNs have the capability to extract specific features from the  raw signals of complex body movements  and   classify  extracted features  into activities such as standing, sitting falling, walking etc. 

Guo and his team used an acoustic grid roughly 40x40 cm in size which  held 256 acoustic receivers in a 16X16 array   and  4 ultrasonic transmitters in the centre. The transmitters  emitted  high frequency sinusoidal acoustic  signals inclined at 45 deg. with  an effective reach  of 4meters.  The human volunteers selected for the experiment varied in height and weight.  They, one at a time, repeatedly performed   activities such as standing, sitting, falling and walking at a distance of 2 meters from the gadget. The ultrasonic sensors  collected  the reflected signals and the  CNN processor did the rest of the work.  Guo et al found  that the accuracy of HAR was 100% for simple static modes such as standing  and sitting  and 97.5% for others.  They also report  that higher the number of sensors and iterations,  higher the  recognition  accuracy for complex activities such as walking and falling. 

Guo and his team are of the opinion that acoustic surveillance is less intrusive of privacy than visual mode. Well, whatever that be,   Eavesdropping 4G has arrived!

REFERENCES:
1.Deep Learning Models for Human Activity Recognition
2.Deep learning for sensor -based activity recognition A survey: Pattern Recognition Letters, vol.119, pp3-11 (2019)
3. Convolutional neural networks for human activity recognition using body-worn sensors
Rueda et al Informatics: 5 (26)  pp (2018)
4. A single feature for human activity recognition using two dimensional acoustic array.
Guo et al, Applied Physics Letters  Vol.114, 214101 (2019)

Resuscitating the Brain: Introducing BrainEx

cerebral circulation
courtsey: wiki BruceBlaus

Healthy brain needs continuous and copious supply of oxygen for functioning. Cerebral circulation ensures that  oxygenated blood is   pumped in  and   oxygen depleted  blood is pumped out. If the oxygen  supply or in other words  circulation is interrupted,  then within seconds brain cells suffer irreparable damage and within minutes this  biological wonder called brain is declared dead. A point of no return. Well that  has been the belief so far.  

Now  Vrselja et al demonstrate that even several hours after death, circulation and cellular functions can be at least  partially revived in mammalian brain.  Sounds eerie indeed. But science fiction had already conceived this plot almost a century ago. Remember the 1925  Russian novel Head of Prof. Dowell  ?.  

But let us get back to  Vrselja et al.   To fully comprehend the significance and complexity of the experiments it is necessary to  highlight the essential features of their investigations. Experiments were conducted on pig brains  procured from USDA approved  food processing facilities within 4 hours post mortem.  Brains were surgically isolated just above the brain stem (medula oblongata), so as to keep the vascular  system intact. The vascular system is necessary to circulate haemoglobin based (but  cell-free)  blood substitute. The entire experimental set-up, blood surrogate, various pumps, monitors, control units and computers, scanners   is referred to as BrainEx. or simply BEx     

32 pig brains 4 hours after disembodiment were carefully scooped out of the skull and hooked on  to BrainEx maintained at 37 deg.C.  First  the research team    established  that  the specially  concocted  blood surrogate enriched with nutrient and medicinal molecules  flowed in and drained out through the major, minor and capillary vessels in the brain. Within an hour sodium and potassium levels registered normal levels. The team could  detect cerebral  metabolic activity as recorded by oxygen and glucose consumption data.   Drugs meant to increase the circulation did exactly the same. Most importantly circulation ensured that  swelling of brain (oedema) is contained and thus  structural integrity of the organ is preserved. The team could keep the system going for a full 6 hours. That meant a total 10 hours after decapitation. During these 10 hours, individual neurones fired at random however  there was no collective or concerted  activity.  Not yet. 

What does this mean and where does this lead us to? Vrselja and team  conclude that mammalian brain is much more resilient than our current assumptions. They are confident that more sophisticated  versions of BEx  could achieve much more.  However  an  ethical dilemma too is looming large. Younger and Hyun both Professors of Bioethics at Case Western Reserve University School of Medicine, Cleveland, Ohio, state that " even now clinicians and bioethicist disagree over how long is long enough for paramedics to keep trying to resuscitate". 

REFERENCES:
1. Professor Dowell's Head by Alexander Belyayev
2. Restoration of brain circulation and cellular functions hours post-mortem: Vrselja et al  
    Nature 18 April 2019 vol.568, pp336-343 
2. Part revived pig brains raise ethical quandaries: Farahany et al Nature  18 April 2019-  
     vol.568,pp299-302
3. Pig brain study could fuel debate around death: Youngner et al Nature 18 April 2019,  
    vol.568, pp302-304

On Alzheimer's Disease

The  French play Le Père by Florian Zeller showcases   Andrei, an old man whose memory is clouded by  Alzheimer's disease(AD).  Le Père received  the 2014  Molière Award for best play. The play has been translated into and staged in several languages. Imaginatively conceived and performed,  audience can empathise with  the vulnerability and frustration of Andrei as his memories fade and/or  at times randomly light up.  We recently had the privilege of  watching it at Ranga Shankara, Bangalore. 

 Courtesy Wikipedia 

Experts say it all happens in the hippocampus,  the seat of  short and long term memory , learning and hence  cognition.   One of the earliest  signs of AD is the  accumulation of   clumps of β-amyloid protein (βA proteins)  in between neurons, and entanglements of tau protein (𝝉 protein) fibrils  within. Both these break up neural communication network, leading to memory loss, cognition deficiency etc.  So naturally the question arises, can we clean up this messy stuff that builds up outside and  inside ? Or better still, can we  find ways to stop its production? Perhaps then we can re-establish  normal neuronal activity and thus restore  memories and cognition?  

Drug companies have invested billions of dollars  to find out a suitable answer. Still over the last 20 years only 4 drugs have hit the market. One novel approach has been  to develop  monoclonal antibodies, which  can loosen up amyloid plaques and lyse them. Though looked promising during  initial stages,  many candidates in this line-up  failed in later stages. In a field where   failure rate of 99.6% is routine, scientists are not disheartened.   Further more, realisation has sunk in that multiplicity of factors could lead to  AD.  And  hence  a single magic drug will not arrest or reverse  the progress of the disease for all.   

Tailpiece:

According to the World Alzheimer Report 2018, currently 50 million people are suffering from AD and by 2050 this number would be 150 million.  If left untreated we might eventually  reach an alarming  situation with majority of the world population afflicted  with  semantic dementia as described by Garcia in One Hundred Years of Solitude.  Macondo  was first robbed off its sleep and then people lost   their vocabulary!  

REFERENCES:
1.World Alzheimer Report 2018
2.The antibody aducanumab reduces Aβ plaques in Alzheimer's disease

3. Aducanumab at Clinical Trials on Alzheimer’s Disease (CTAD)4.Biogen Halts Promising Alzheimer's Drug Trials Due to Disappointing Results
5.Initiation of Phase III clinical trial of BAN2401 in early Alzheimer's disease
6.Semantic Dementia in One Hundred Years of Solitude

Seeing (in the dark) is believing.

Perhaps because we humans were destined to be  diurnal and not  nocturnal species, evolution  didn't  equip us with night vision. Instead we received cone cells for  color vision,  ideal for enjoying the wide spectrum of colours in broad day light.  On the other hand  nocturnals  developed a thin membrane tapetum lucidum,  behind the retina which functions as a rear light reflector  enabling vision in low light and  during night.  None the less we humans know how to jump over natural barriers with the help of technology; thus we have   extended our working hours into the darkness of night and peeped through night vision goggles and infrared cameras to see what cannot be seen with naked eye.  Night vision devices are available in a wide range of  prices/functionalities.  

Because we  are bent upon following knowledge like a sinking star, beyond the utmost bounds of human thought,  we have breeched another barrier. No  not for ourselves, but  for rodents. A research team led by  neuroscientist Tian Xue at the University of Science and Technology, Hefei, China has given the power of  night vision to mice through an  ocular injection  of nanoparticles.  Xue says it started as a crazy idea.    Nano particles of ytterbium and erbium, two metals that belong to the rare earth group caught the attention of the team.  Ytterbium can absorb infrared light and transfer it to neighbouring erbium which in turn responds with the emission of  green light. The team tagged these nanoparticles with a protein called ConA which preferentially attaches to the surface of photoreceptor cells in the retina  This concoction was injected into the rat's retina.  ConA latches on the retinal cells.   Once safely anchored there,  ytterbium and erbium nanoparticles work in unison converting infra red signals to green light which retina picks up, converts to electrical signals and sends to visual cortex for processing and decoding.  Xue and team ran a series of experiments to monitor the behaviour  of rats in dark and infrared light lit chambers. These experiments  proved conclusively  that the injected mice  could actually see in the dark. 

A concept has been proven and a patent has been applied for.  Xue et al are confident  of the  civilian and military applications of the technique.    It is indeed  debatable whether night vision would improve  the quality of human  life;  but certainly ghosts would scramble  to  hide in plain sight. 

Reference 

1. Mammalian Near-Infrared Image Vision through Injectable and Self-Powered Retinal Nanoantennae.                                                                                                                   Cell, 2019; DOI: 10.1016/j.cell.2019.01.038 Yuqian Ma, Jin Bao, Yuanwei Zhang, Zhanjun Li, Xiangyu Zhou, Changlin Wan, Ling Huang, Yang Zhao, Gang Han, Tian Xue.

2. Researchers enable "Super Mice" to see Near-Infrared Light

Dance Lessons for Humanoids

The website of Boston Dynamics features several videos on the acrobatics of Atlas, the Humanoid. With a height of 1.8 meters supporting a weight of 75 Kg, Atlas has a total of 28 joints and runs on battery power. It walks, runs, jumps, somersaults. Kengoro, the latest humanoid  from University of Tokyo, is 1.7 meters tall and 56kg in weight, and is fitted with  160 motors for anatomical flexibility. Kengoro has a unique cooling system which allows it to sweat.  But both sorely miss out on  aesthetics.     Perhaps  aesthetics was remote from the minds of their designers. 

Atlas,
Courtesy: Wikipedia

Sophia,
Courtesy: Wikipedia

Sophia, the humanoid designed by Hanson Robotics is a study in contrast. Modelled on the yesteryear British actress Audrey Hepburn, Sophia has more humanlike features.  Media is brimming  with her social capabilities.  The Hanson website states that  Sophia, (which means wisdom in Greek) is designed to achieve  meaningful   collaboration between humans and Artificial Intelligence.  Miim, very Japanese in her attributes is petite, with  just 158 centimeters in height and 43 kg weight. She can walk, speak and even dance.  She is the creation of National Institute of Advanced Industrial Science and Technology (AIST), Tokyo.

Current projections estimate that by 2025, robots will execute  25% of manual labor. However socially assistive robots in health and services industries need to be endowed with more human-like traits. One of the areas design engineers need to focus is robotic movements. The movements of the robot/humanoid are  jerky and abrupt.   It is necessary to impart more fluidity to these movements.  Muscles which smoothly contract and stretch allow humans to move gradually, gracefully.   Briefly reviewing the trends in the field  of Humanoid Designs, Amy La Viers, ( Director of Robotics, Automation and Dance Lab at the University of Illinois at Urbana-Champaign) observes that "The diversity of robot movements needs to be expanded  to include variable and complex motions."  She suggests that  roboticists must work in collaboration with dancers and choreographers to refine the movements of the robots.

Reference 
1. How long until robots rule the world?
2. A Music-Driven Dance System of Humanoid Robots

3. How to make a Humanoid Robot Dance?
4. Make Robot motions natural : La Viers, Nature pp 422-424 Vol.565 January 2019

A Twist: At a Magic Angle

Graphene Artist's representation
: Courtesy Wiki

It is unbelievably true. A small twist,  hardly 1.1degree and lo and behold you have a superconducting material. Scientists have been trying tirelessly to tame the phenomenon of superconductivity and to make materials that exhibit this property at room temperature.  Not that we are anywhere near room temperature  as yet, but something amazing has caught the attention of scientists. And it is about graphene  sheets.

Everything about Graphene has been unconventional,  from the very beginning. Graphene is a mono layer of carbon atoms  arranged in a hexagonal format. Scientists   Geim and Novosilov  (who later received the Nobel Prize in Physics 2010) were the first to  peel off such layers from a graphite lump using a scotch tape. 

A team of scientists 12 years ago demonstrated that  graphene  bilayer could be turned into  a tunable  semiconductor when a small electric voltage is  applied  across the sheet. 

And now MIT Professor  Pablo Jarillo-Herrero and his team of students are flying high as they  demonstrate that a slightly misaligned pair of graphene sheets, cooled to almost absolute zero, could switch its behaviour  between an insulator and an unconventional  Superconductor, when a voltage is applied.  

A slight twist of 1.1 degree- as simple as that?, Scientists all over the world are wondering.  At the recent Annual Meeting of the American Physical Society, when Jarillo-Herrero presented his results, the lecture hall was overflowing; audience spilled over and  stood wherever they could, to  hear him talk.  The theory behind the phenomenon remains to be worked with the rigour it demands. However the general assumption is that the 1.1 degree twist  drastically alters  the electronic properties of the ensemble. Graphene is a pure clean system and scientists are confident that it  is an easier system to study than those complicated mixed oxides which have earlier exhibited such property. 

Twistronics is the buzz word now.  As one scientist puts it everyone is taking their favourite thing and twisting it with their other favourite thing, all in the hope of striking  it rich with superconductivity.


References:
1. Insulator or Superconductor? MIT News 
2. Superconductivity with a twist: Gibney,E.; Nature 2019, 565 pp15-18
2. Y. Cao et al., “Correlated insulator behaviour at half-filling in magic-angle graphene superlattices,” Nature 556, 80 (2018).
3. Y. Cao et al., “Unconventional superconductivity in magic-angle graphene superlattices,” Nature 556, 43 (2018).
4.Carr et al., Twistronics : Manipulating the Electronic Properties of Two-dimensional Layered structures through their Twist Angle  APS March Meeting 2017 Abstract E33.003