Computers Math

Artificial intelligence (AI) experts at the University of Massachusetts Amherst and the Baylor College of Medicine report that they have successfully addressed what they call a “major, long-standing obstacle to increasing AI capabilities” by drawing inspiration from a human brain memory mechanism known as “replay.”
First author and postdoctoral researcher Gido van de Ven and principal investigator Andreas Tolias at Baylor, with Hava Siegelmann at UMass Amherst, write in Nature Communications that they have developed a new method to protect — “surprisingly efficiently” — deep neural networks from “catastrophic forgetting” — upon learning new lessons, the networks forget what they had learned before.
Siegelmann and colleagues point out that deep neural networks are the main drivers behind recent AI advances, but progress is held back by this forgetting.
They write, “One solution would be to store previously encountered examples and revisit them when learning something new. Although such ‘replay’ or ‘rehearsal’ solves catastrophic forgetting,” they add, “constantly retraining on all previously learned tasks is highly inefficient and the amount of data that would have to be stored becomes unmanageable quickly.”
Unlike AI neural networks, humans are able to continuously accumulate information throughout their life, building on earlier lessons. An important mechanism in the brain believed to protect memories against forgetting is the replay of neuronal activity patterns representing those memories, they explain.
Siegelmann says the team’s major insight is in “recognizing that replay in the brain does not store data.” Rather, “the brain generates representations of memories at a high, more abstract level with no need to generate detailed memories.” Inspired by this, she and colleagues created an artificial brain-like replay, in which no data is stored. Instead, like the brain, the network generates high-level representations of what it has seen before.
The “abstract generative brain replay” proved extremely efficient, and the team showed that replaying just a few generated representations is sufficient to remember older memories while learning new ones. Generative replay not only prevents catastrophic forgetting and provides a new, more streamlined path for system learning, it allows the system to generalize learning from one situation to another, they state.
For example, “if our network with generative replay first learns to separate cats from dogs, and then to separate bears from foxes, it will also tell cats from foxes without specifically being trained to do so. And notably, the more the system learns, the better it becomes at learning new tasks,” says van de Ven.
He and colleagues write, “We propose a new, brain-inspired variant of replay in which internal or hidden representations are replayed that are generated by the network’s own, context-modulated feedback connections. Our method achieves state-of-the-art performance on challenging continual learning benchmarks without storing data, and it provides a novel model for abstract level replay in the brain.”
Van de Ven says, “Our method makes several interesting predictions about the way replay might contribute to memory consolidation in the brain. We are already running an experiment to test some of these predictions.”

Story Source:
Materials provided by University of Massachusetts Amherst. Note: Content may be edited for style and length. More

High-speed cameras can take pictures in quick succession. This makes them useful for visualizing ultrafast dynamic phenomena, such as femtosecond laser ablation for precise machining and manufacturing processes, fast ignition for nuclear fusion energy systems, shock-wave interactions in living cells, and certain chemical reactions.
Among the various parameters in photography, the sequential imaging of microscopic ultrafast dynamic processes requires high frame rates and high spatial and temporal resolutions. In current imaging systems, these characteristics are in a tradeoff with one another.
However, scientists at Shenzhen University, China, have recently developed an all-optical ultrafast imaging system with high spatial and temporal resolutions, as well as a high frame rate. Because the method is all-optical, it’s free from the bottlenecks that arise from scanning with mechanical and electronic components.
Their design focuses on non-collinear optical parametric amplifiers (OPAs). An OPA is a crystal that, when simultaneously irradiated with a desired signal light beam and a higher-frequency pump light beam, amplifies the signal beam and produces another light beam known as an idler. Because the crystal used in this study is non-collinear, the idler is fired in a different direction from that of the signal beam. But how is such a device useful in a high-speed imaging system?
The answer lies in cascading OPAs. The information of the target, contained in the signal beam, is mapped onto the idler beam by the OPA while the pump beam is active. Because the idler moves in a different direction, it can be captured using a conventional charge-coupled device (CCD) camera “set to the side” while the signal beam moves toward the next stage in the OPA cascade.
Just like how water would descend in a waterfall, the signal beam reaches the subsequent OPA, and the pump beam generated from the same laser source activates it; except now, a delay line makes the pump beam arrive later, causing the CCD camera next to the OPA in the second stage to take a picture later. Through a cascade of four OPAs with four associated CCD cameras and four different delay lines for the pump laser, the scientists created a system that can take four pictures in extremely quick succession.
The speed of capturing consecutive pictures is limited by how small the difference between two laser delay lines can be. In this regard, this system achieved an effective frame rate of 15 trillion frames per second — a record shutter speed for high-spatial-resolution cameras. Conversely, the temporal resolution depends on the duration of the laser pulses triggering the OPAs and generating the idler signals. In this case, the pulse width was 50 fs (fifty millionths of a nanosecond). Coupled with the incredibly fast frame rate, this method is able to observe ultrafast physical phenomena, such as an air plasma grating and a rotating optical field spinning at 10 trillion radians per second.
According to Anatoly Zayats, Co-Editor-in-Chief of Advanced Photonics, “The team at Shenzhen University has demonstrated ultrafast photographic imaging with the record fastest shutter speed. This research opens up new opportunities for studies of ultrafast processes in various fields.”
This imaging method has scope for improvement but could easily become a new microscopy technique. Future research will unlock the potential of this approach to give us a clearer picture of ultrafast transient phenomena. More

Cornell University systems engineers examined data from a busy New York state food bank and, using a new algorithm, found ways to better distribute and allocate food, and elevate nutrition among its patrons in the process.
“In order to serve thousands of people and combat food insecurity, our algorithm helps food banks manage their food resources more efficiently — and patrons get more nutrition,” said lead researcher Faisal Alkaabneh, Cornell’s first doctoral graduate in systems engineering.
Alkaabneh and his adviser, Oliver Gao, professor of civil and environmental engineering, are co-authors of “A Unified Framework for Efficient, Effective and Fair Resource Allocation by Food Banks Using an Approximate Dynamic Programming Approach,” published in the journal Omega.
The researchers reviewed data of the Food Bank of the Southern Tier, which serves six counties in upstate New York. In 2019, the food bank distributed 10.9 million meals to about 21,700 people each week. Nearly 19% of its patrons are seniors and about 41% are children, according to the group’s data.
Last year, the food bank distributed 2.8 million pounds of fresh fruit through 157 partner agencies, and moved about 3.4 million pounds of food through local mobile pantries.
The algorithm Gao and his team used to determine how to allocate several food categories efficiently, based upon pantry requests, demonstrated a 7.73% improvement in efficiency from 2018 to 2019, compared to standard food bank allocation practices. Their calculations also showed a 3% improvement in nutrition using a wider variety of food, Alkaabneh said.
“We hope our research is used as a baseline model for food banks improving practices,” Gao said. “and boosting nutrition and policies to help people at risk for hunger.”

Story Source:
Materials provided by Cornell University. Original written by Blaine Friedlander. Note: Content may be edited for style and length. More

It is better to invest in measures that make it easier for women to visit a doctor during pregnancy than measures to repair birth injuries. This is the conclusion from two mathematicians at LiU, using Uganda as an example.
A fistula is a connection between the vagina and bladder or between the vagina and the rectum. It can arise in women during prolonged childbirth or as a result of violent rape. The connection causes incontinence, in which the patient cannot control either urine or faeces, which in turn leads to several further medical problems and to major physical, mental and social suffering. The medical care system in Uganda is well-developed, particularly in metropolitan areas, but despite this the occurrence of fistulas during childbirth is among the highest in Africa. It has been calculated that between 1.63 and 2.25 percent of women of childbearing age, 15-49 years, are affected.
Betty Nannyonga, postdoc at LiU who also works at Makerere University in Kampala, Uganda, and Martin Singull, associate professor in mathematical statistics at the Department of Mathematics at LiU, have published an article in the scientific journal PLOS ONE that demonstrates how the available resources can be put to best use.
“We have tried to construct a mathematical model to show how to prevent (obstetric) fistulas in women during prolonged childbirth. This is hugely important in a country such as Uganda,” says Martin Singull.
The study analysed data from the Uganda Demographic Health Survey 2016. This survey collected information from 18,506 women aged 15-49 years and living in 15 regions in Uganda. Some special clinics, known as “fistula camps,” have been set up in recent years to provide surgery for affected women. Data from two of these, in different parts of the country, were also included in the study.
The research found that significantly fewer women have received surgery at the clinics than expected.

advertisement

“Our results show that Uganda has a huge backlog of women who should be treated for fistula. In one of the regions we looked at, we found that for each woman who had undergone an operation, at least eight more should have received care,” says Betty Nannyonga.
The researchers have looked at the relationship between the resources required to provide surgery for the women after being injured and the corresponding resources used to give the women access to professional care during the pregnancy, which includes the availability of delivery by Caesarean section if required. Another chilling statistic that must be considered is that not only do the women suffer from fistula, but that the baby survives in only 9% of such cases.
The mathematical models demonstrate that the number of women who suffer from fistula decreases most rapidly if the resources are put into preventive maternity care, and making it possible for the women to give birth in hospital.
The authors point out, however, that several difficulties contribute to the relatively high occurrence of fistula.
“Even if professional healthcare and medical care are available in Uganda, most women do not enjoy good maternity care during their pregnancy. In some cases, this is because the distance to the healthcare providers is too far. Other reasons are the women do not have the money needed, or that they require permission from their husbands. It won’t do any good to invest money in health centres if the women don’t attend,” says Betty Nannyonga.
The regions differ considerably in how they provide care. It is twice as likely that a woman in a town will see a doctor than it is for someone who lives in rural areas. Another factor is education: those with upper secondary education are twice as likely as those without, and having higher education than upper secondary increases the probability by a further factor of two. Social status also plays a major role: the probability of seeing a doctor on some occasion during the pregnancy is directly linked to income.
One positive trend shown by the statistics is an increase in the fraction of women who receive professional care during the delivery itself (although the figures are for only those cases in which a child is born alive). This has risen from 37% in the period before 2001, to 42% in 2006, 58% in 2011 and 74% in 2016.
“Our results show that professional care and surgery by themselves cannot prevent all cases of fistula: other measures will be needed,” concludes Betty Nannyonga.

Story Source:
Materials provided by Linköping University. Original written by Monica Westman Svenselius. Note: Content may be edited for style and length. More

Researchers at Helmholtz Zentrum München and the Technical University of Munich (TUM) have developed the world’s smallest ultrasound detector. It is based on miniaturized photonic circuits on top of a silicon chip. With a size 100 times smaller than an average human hair, the new detector can visualize features that are much smaller than previously possible, leading to what is known as super-resolution imaging.
Since the development of medical ultrasound imaging in the 1950s, the core detection technology of ultrasound waves has primarily focused on using piezoelectric detectors, which convert the pressure from ultrasound waves into electric voltage. The imaging resolution achieved with ultrasound depends on the size of the piezoelectric detector employed. Reducing this size leads to higher resolution and can offer smaller, densely packed one or two dimensional ultrasound arrays with improved ability to discriminate features in the imaged tissue or material. However, further reducing the size of piezoelectric detectors impairs their sensitivity dramatically, making them unusable for practical application.
Using computer chip technology to create an optical ultrasound detector
Silicon photonics technology is widely used to miniaturize optical components and densely pack them on the small surface of a silicon chip. While silicon does not exhibit any piezoelectricity, its ability to confine light in dimensions smaller than the optical wavelength has already been widely exploited for the development of miniaturized photonic circuits.
Researchers at Helmholtz Zentrum Mu?nchen and TUM capitalized on the advantages of those miniaturized photonic circuits and built the world’s smallest ultrasound detector: the silicon waveguide-etalon detector, or SWED. Instead of recording voltage from piezoelectric crystals, SWED monitors changes in light intensity propagating through the miniaturized photonic circuits.
“This is the first time that a detector smaller than the size of a blood cell is used to detect ultrasound using the silicon photonics technology,” says Rami Shnaiderman, developer of SWED. “If a piezoelectric detector was miniaturized to the scale of SWED, it would be 100 million times less sensitive.”
Super-resolution imaging

advertisement

“The degree to which we were we able to miniaturize the new detector while retaining high sensitivity due to the use of silicon photonics was breathtaking,” says Prof. Vasilis Ntziachristos, lead of the research team. The SWED size is about half a micron (=0,0005 millimeters). This size corresponds to an area that is at least 10,000 times smaller than the smallest piezoelectric detectors employed in clinical imaging applications. The SWED is also up to 200 times smaller than the ultrasound wavelength employed, which means that it can be used to visualize features that are smaller than one micrometer, leading to what is called super-resolution imaging.
Inexpensive and powerful
As the technology capitalizes on the robustness and easy manufacturability of the silicon platform, large numbers of detectors can be produced at a small fraction of the cost of piezoelectric detectors, making mass production feasible. This is important for developing a number of different detection applications based on ultrasound waves. “We will continue to optimize every parameter of this technology — the sensitivity, the integration of SWED in large arrays, and its implementation in hand-held devices and endoscopes,” adds Shnaiderman.
Future development and applications
“The detector was originally developed to propel the performance of optoacoustic imaging, which is a major focus of our research at Helmholtz Zentrum München and TUM. However, we now foresee applications in a broader field of sensing and imaging,” says Ntziachristos.
While the researchers are primarily aiming for applications in clinical diagnostics and basic biomedical research, industrial applications may also benefit from the new technology. The increased imaging resolution may lead to studying ultra-fine details in tissues and materials. A first line of investigation involves super-resolution optoacoustic (photoacoustic) imaging of cells and micro-vasculature in tissues, but the SWED could be also used to study fundamental properties of ultrasonic waves and their interactions with matter on a scale that was not possible before. More

A medical robotic hand could allow doctors to more accurately diagnose and treat people from halfway around the world, but currently available technologies aren’t good enough to match the in-person experience.
Researchers report in Science Advances that they have designed and produced a smart electronic skin and a medical robotic hand capable of assessing vital diagnostic data by using a newly invented rubbery semiconductor with high carrier mobility.
Cunjiang Yu, Bill D. Cook Associate Professor of Mechanical Engineering at the University of Houston and corresponding author for the work, said the rubbery semiconductor material also can be easily scaled for manufacturing, based upon assembly at the interface of air and water.
That interfacial assembly and the rubbery electronic devices described in the paper suggest a pathway toward soft, stretchy rubbery electronics and integrated systems that mimic the mechanical softness of biological tissues, suitable for a variety of emerging applications, said Yu, who also is a principal investigator at the Texas Center for Superconductivity at UH.
The smart skin and medical robotic hand are just two potential applications, created by the researchers to illustrate the discovery’s utility.
In addition to Yu, authors on the paper include Ying-Shi Guan, Anish Thukral, Kyoseung Sim, Xu Wang, Yongcao Zhang, Faheem Ershad, Zhoulyu Rao, Fengjiao Pan and Peng Wang, all of whom are affiliated with UH. Co-authors Jianliang Xiao and Shun Zhang are affiliated with the University of Colorado.
Traditional semiconductors are brittle, and using them in otherwise stretchable electronics has required special mechanical accommodations. Previous stretchable semiconductors have had drawbacks of their own, including low carrier mobility — the speed at which charge carriers can move through a material — and complicated fabrication requirements.
Yu and collaborators last year reported that adding minute amounts of metallic carbon nanotubes to the rubbery semiconductor of P3HT — polydimethylsiloxane composite — improves carrier mobility, which governs the performances of semiconductor transistors.
Yu said the new scalable manufacturing method for these high performance stretchable semiconducting nanofilms and the development of fully rubbery transistors represent a significant step forward.
The production is simple, he said. A commercially available semiconductor material is dissolved in a solution and dropped on water, where it spreads; the chemical solvent evaporates from the solution, resulting in improved semiconductor properties.
It is a new way to create the high quality composite films, he said, allowing for consistent production of fully rubbery semiconductors.
Electrical performance is retained even when the semiconductor is stretched by 50%, the researchers reported. Yu said the ability to stretch the rubbery electronics by 50% without degrading the performance is a notable advance. Human skin, he said, can be stretched only about 30% without tearing.

Story Source:
Materials provided by University of Houston. Original written by Jeannie Kever. Note: Content may be edited for style and length. More

Artificial intelligence researchers at North Carolina State University have improved the performance of deep neural networks by combining feature normalization and feature attention modules into a single module that they call attentive normalization (AN). The hybrid module improves the accuracy of the system significantly, while using negligible extra computational power.
“Feature normalization is a crucial element of training deep neural networks, and feature attention is equally important for helping networks highlight which features learned from raw data are most important for accomplishing a given task,” says Tianfu Wu, corresponding author of a paper on the work and an assistant professor of electrical and computer engineering at NC State. “But they have mostly been treated separately. We found that combining them made them more efficient and effective.”
To test their AN module, the researchers plugged it into four of the most widely used neural network architectures: ResNets, DenseNets, MobileNetsV2 and AOGNets. They then tested the networks against two industry standard benchmarks: the ImageNet-1000 classification benchmark and the MS-COCO 2017 object detection and instance segmentation benchmark.
“We found that AN improved performance for all four architectures on both benchmarks,” Wu says. “For example, top-1 accuracy in the ImageNet-1000 improved by between 0.5% and 2.7%. And Average Precision (AP) accuracy increased by up to 1.8% for bounding box and 2.2% for semantic mask in MS-COCO.
“Another advantage of AN is that it facilitates better transfer learning between different domains,” Wu says. “For example, from image classification in ImageNet to object detection and semantic segmentation in MS-COCO. This is illustrated by the performance improvement in the MS-COCO benchmark, which was obtained by fine-tuning ImageNet-pretrained deep neural networks in MS-COCO, a common workflow in state-of-the-art computer vision.
“We have released the source code and hope our AN will lead to better integrative design of deep neural networks.”

Story Source:
Materials provided by North Carolina State University. Note: Content may be edited for style and length. More

Researchers have developed an advanced spectrometer that can acquire data with exceptionally high speed. The new spectrometer could be useful for a variety of applications including remote sensing, real-time biological imaging and machine vision.
Spectrometers measure the color of light absorbed or emitted from a substance. However, using such systems for complex and detailed measurement typically requires long data acquisition times.
“Our new system can measure a spectrum in mere microseconds,” said research team leader Scott B. Papp from the National Institute of Standards and Technology and the University of Colorado, Boulder. “This means it could be used for chemical studies in the dynamic environment of power plants or jet engines, for quality control of pharmaceuticals or semiconductors flying by on a production line, or for video imaging of biological samples.”
In The Optical Society (OSA) journal Optics Express, lead author David R. Carlson and colleagues Daniel D. Hickstein and Papp report the first dual-comb spectrometer with a pulse repetition rate of 10 gigahertz. They demonstrate it by carrying out spectroscopy experiments on pressurized gases and semiconductor wafers.
“Frequency combs are already known to be useful for spectroscopy,” said Carlson. “Our research is focused on building new, high-speed frequency combs that can make a spectrometer that operates hundreds of times faster than current technologies.”
Getting data faster
Dual-comb spectroscopy uses two optical sources, known as optical frequency combs that emit a spectrum of colors — or frequencies — perfectly spaced like the teeth on a comb. Frequency combs are useful for spectroscopy because they provide access to a wide range of colors that can be used to distinguish various substances.

advertisement

To create a dual-comb spectroscopy system with extremely fast acquisition and a wide range of colors, the researchers brought together techniques from several different disciplines, including nanofabrication, microwave electronics, spectroscopy and microscopy.
The frequency combs in the new system use an optical modulator driven by an electronic signal to carve a continuous laser beam into a sequence of very short pulses. These pulses of light pass through nanophotonic nonlinear waveguides on a microchip, which generates many colors of light simultaneously. This multi-color output, known as a supercontinuum, can then be used to make precise spectroscopy measurements of solids, liquids and gases.
The chip-based nanophotonic nonlinear waveguides were a key component in this new system. These channels confine light within structures that are a centimeter long but only nanometers wide. Their small size and low light losses combined with the properties of the material they are made from allow them to convert light from one wavelength to another very efficiently to create the supercontinuum.
“The frequency comb source itself is also unique compared to most other dual-comb systems because it is generated by carving a continuous laser beam into pulses with an electro-optic modulator,” said Carlson. “This means the reliability and tunability of the laser can be exceptionally high across a wide range of operating conditions, an important feature when looking at future applications outside of a laboratory environment.”
Analyzing gases and solids
To demonstrate the versatility of the new dual-comb spectrometer, the researchers used it to perform linear absorption spectroscopy on gases of different pressure. They also operated it in a slightly different configuration to perform the advanced analytical technique known as nonlinear Raman spectroscopy on semiconductor materials. Nonlinear Raman spectroscopy, which uses pulses of light to characterize the vibrations of molecules in a sample, has not previously been performed using an electro-optic frequency comb.

advertisement

The high data acquisition speeds that are possible with electro-optic combs operating at gigahertz pulse rates are ideal for making spectroscopy measurements of fast and non-repeatable events.
“It may be possible to analyze and capture the chemical signatures during an explosion or combustion event,” said Carlson. “Similarly, in biological imaging the ability to create images in real time of living tissues without requiring chemical labeling would be immensely valuable to biological researchers.”
The researchers are now working to improve the system’s performance to make it practical for applications like real-time biological imaging and to simplify and shrink the experimental setup so that it could be operated outside of the lab.

Story Source:
Materials provided by The Optical Society. Note: Content may be edited for style and length. More