Aurelia Butler

Compared to standard machine learning models, deep learning models are largely superior at discerning patterns and discriminative features in brain imaging, despite being more complex in their architecture, according to a new study in Nature Communications led by Georgia State University.
Advanced biomedical technologies such as structural and functional magnetic resonance imaging (MRI and fMRI) or genomic sequencing have produced an enormous volume of data about the human body. By extracting patterns from this information, scientists can glean new insights into health and disease. This is a challenging task, however, given the complexity of the data and the fact that the relationships among types of data are poorly understood.
Deep learning, built on advanced neural networks, can characterize these relationships by combining and analyzing data from many sources. At the Center for Translational Research in Neuroimaging and Data Science (TReNDS), Georgia State researchers are using deep learning to learn more about how mental illness and other disorders affect the brain.
Although deep learning models have been used to solve problems and answer questions in a number of different fields, some experts remain skeptical. Recent critical commentaries have unfavorably compared deep learning with standard machine learning approaches for analyzing brain imaging data.
However, as demonstrated in the study, these conclusions are often based on pre-processed input that deprive deep learning of its main advantage — the ability to learn from the data with little to no preprocessing. Anees Abrol, research scientist at TReNDS and the lead author on the paper, compared representative models from classical machine learning and deep learning, and found that if trained properly, the deep-learning methods have the potential to offer substantially better results, generating superior representations for characterizing the human brain.
“We compared these models side-by-side, observing statistical protocols so everything is apples to apples. And we show that deep learning models perform better, as expected,” said co-author Sergey Plis, director of machine learning at TReNDS and associate professor of computer science.

advertisement

Plis said there are some cases where standard machine learning can outperform deep learning. For example, diagnostic algorithms that plug in single-number measurements such as a patient’s body temperature or whether the patient smokes cigarettes would work better using classical machine learning approaches.
“If your application involves analyzing images or if it involves a large array of data that can’t really be distilled into a simple measurement without losing information, deep learning can help,” Plis said.. “These models are made for really complex problems that require bringing in a lot of experience and intuition.”
The downside of deep learning models is they are “data hungry” at the outset and must be trained on lots of information. But once these models are trained, said co-author Vince Calhoun, director of TReNDS and Distinguished University Professor of Psychology, they are just as effective at analyzing reams of complex data as they are at answering simple questions.
“Interestingly, in our study we looked at sample sizes from 100 to 10,000 and in all cases the deep learning approaches were doing better,” he said.
Another advantage is that scientists can reverse analyze deep-learning models to understand how they are reaching conclusions about the data. As the published study shows, the trained deep learning models learn to identify meaningful brain biomarkers.
“These models are learning on their own, so we can uncover the defining characteristics that they’re looking into that allows them to be accurate,” Abrol said. “We can check the data points a model is analyzing and then compare it to the literature to see what the model has found outside of where we told it to look.”
The researchers envision that deep learning models are capable of extracting explanations and representations not already known to the field and act as an aid in growing our knowledge of how the human brain functions. They conclude that although more research is needed to find and address weaknesses of deep-learning models, from a mathematical point of view, it’s clear these models outperform standard machine learning models in many settings.
“Deep learning’s promise perhaps still outweighs its current usefulness to neuroimaging, but we are seeing a lot of real potential for these techniques,” Plis said. More

Physicists at Washington University in St. Louis have discovered how to locally add electrical charge to an atomically thin graphene device by layering flakes of another thin material, alpha-RuCl3, on top of it.
A paper published in the journal Nano Letters describes the charge transfer process in detail. Gaining control of the flow of electrical current through atomically thin materials is important to potential future applications in photovoltaics or computing.
“In my field, where we study van der Waals heterostructures made by custom-stacking atomically thin materials together, we typically control charge by applying electric fields to the devices,” said Erik Henriksen, assistant professor of physics in Arts & Sciences and corresponding author of the new study, along with Ken Burch at Boston College. “But here it now appears we can just add layers of RuCl33. It soaks up a fixed amount of electrons, allowing us to make ‘permanent’ charge transfers that don’t require the external electric field.”
Jesse Balgley, a graduate student in Henriksen’s laboratory at Washington University, is second author of the study. Li Yang, professor of physics, and his graduate student Xiaobo Lu, also both at Washington University, helped with computational work and calculations, and are also co-authors.
Physicists who study condensed matter are intrigued by alpha-RuCl3 because they would like to exploit certain of its antiferromagnetic properties for quantum spin liquids.
In this new study, the scientists report that alpha-RuCl3 is able to transfer charge to several different types of materials — not just graphene, Henriksen’s personal favorite.
They also found that they only needed to place a single layer of alpha-RuCl3 on top of their devices to create and transfer charge. The process still works, even if the scientists slip a thin sheet of an electrically insulating material between the RuCl3 and the graphene.
“We can control how much charge flows in by varying the thickness of the insulator,” Henriksen said. “Also, we are able to physically and spatially separate the source of charge from where it goes — this is called modulation doping.”
Adding charge to a quantum spin liquid is one mechanism thought to underlie the physics of high-temperature superconductivity.
“Anytime you do this, it could get exciting,” Henriksen said. “And usually you have to add atoms to bulk materials, which causes lots of disorder. But here, the charge flows right in, no need to change the chemical structure, so it’s a ‘clean’ way to add charge.”

Story Source:
Materials provided by Washington University in St. Louis. Original written by Talia Ogliore. Note: Content may be edited for style and length. More

There are things in life that can be predicted reasonably well. The tides rise and fall. The moon waxes and wanes. A billiard ball bounces around a table according to orderly geometry.
And then there are things that defy easy prediction: The hurricane that changes direction without warning. The splashing of water in a fountain. The graceful disorder of branches growing from a tree.
These phenomena and others like them can be described as chaotic systems, and are notable for exhibiting behavior that is predictable at first, but grows increasingly random with time.
Because of the large role that chaotic systems play in the world around us, scientists and mathematicians have long sought to better understand them. Now, Caltech’s Lihong Wang, the Bren Professor in the Andrew and Peggy Cherng department of Medical Engineering, has developed a new tool that might help in this quest.
In the latest issue of Science Advances, Wang describes how he has used an ultrafast camera of his own design that recorded video at one billion frames per second to observe the movement of laser light in a chamber specially designed to induce chaotic reflections.
“Some cavities are non-chaotic, so the path the light takes is predictable,” Wang says. But in the current work, he and his colleagues have used that ultrafast camera as a tool to study a chaotic cavity, “in which the light takes a different path every time we repeat the experiment.”
The camera makes use of a technology called compressed ultrafast photography (CUP), which Wang has demonstrated in other research to be capable of speeds as fast as 70 trillion frames per second. The speed at which a CUP camera takes video makes it capable of seeing light — the fastest thing in the universe — as it travels.

advertisement

But CUP cameras have another feature that make them uniquely suited for studying chaotic systems. Unlike a traditional camera that shoots one frame of video at a time, a CUP camera essentially shoots all of its frames at once. This allows the camera to capture the entirety of a laser beam’s chaotic path through the chamber all in one go.
That matters because in a chaotic system, the behavior is different every time. If the camera only captured part of the action, the behavior that was not recorded could never be studied, because it would never occur in exactly the same way again. It would be like trying to photograph a bird, but with a camera that can only capture one body part at a time; furthermore, every time the bird landed near you, it would be a different species. Although you could try to assemble all your photos into one composite bird image, that cobbled-together bird would have the beak of a crow, the neck of a stork, the wings of a duck, the tail of a hawk, and the legs of a chicken. Not exactly useful.
Wang says that the ability of his CUP camera to capture the chaotic movement of light may breathe new life into the study of optical chaos, which has applications in physics, communications, and cryptography.
“It was a really hot field some time ago, but it’s died down, maybe because we didn’t have the tools we needed,” he says. “The experimentalists lost interest because they couldn’t do the experiments, and the theoreticians lost interest because they couldn’t validate their theories experimentally. This was a fun demonstration to show people in that field that they finally have an experimental tool.”
The paper describing the research, titled “Real-time observation and control of optical chaos,” appears in the January 13 issue of Science Advances. Co-authors are Linran Fan, formerly of Caltech, now an assistant professor at Wyant College of Optical Sciences at the University of Arizona; and Xiaodong Yan and Han Wang, of the University of Southern California.
Funding for the research was provided by the Army Research Office Young Investigator Program, the Air Force Office of Scientific Research, the National Science Foundation, and the National Institutes of Health. More

Scientists tame photon-magnon interaction.
Working with theorists in the University of Chicago’s Pritzker School of Molecular Engineering, researchers in the U.S. Department of Energy’s (DOE) Argonne National Laboratory have achieved a scientific control that is a first of its kind. They demonstrated a novel approach that allows real-time control of the interactions between microwave photons and magnons, potentially leading to advances in electronic devices and quantum signal processing.
Microwave photons are elementary particles forming the electromagnetic waves that we use for wireless communications. On the other hand, magnons are the elementary particles forming what scientists call “spin waves” — wave-like disturbances in an ordered array of microscopic aligned spins that can occur in certain magnetic materials.
Microwave photon-magnon interaction has emerged in recent years as a promising platform for both classical and quantum information processing. Yet, this interaction had proved impossible to manipulate in real time, until now.
“Before our discovery, controlling the photon-magnon interaction was like shooting an arrow into the air,” said Xufeng Zhang, an assistant scientist in the Center for Nanoscale Materials, a DOE User Facility at Argonne, and the corresponding author of this work. “One has no control at all over that arrow once in flight.”
The team’s discovery has changed that. “Now, it is more like flying a drone, where we can guide and control its flight electronically,” said Zhang.
By smart engineering, the team employs an electrical signal to periodically alter the magnon vibrational frequency and thereby induce effective magnon-photon interaction. The result is a first-ever microwave-magnonic device with on-demand tunability.
The team’s device can control the strength of the photon-magnon interaction at any point as information is being transferred between photons and magnons. It can even completely turn the interaction on and off. With this tuning capability, scientists can process and manipulate information in ways that far surpass present-day hybrid magnonic devices.
“Researchers have been searching for a way to control this interaction for the past few years,” noted Zhang. The team’s discovery opens a new direction for magnon-based signal processing and should lead to electronic devices with new capabilities. It may also enable important applications for quantum signal processing, where microwave-magnonic interactions are being explored as a promising candidate for transferring information between different quantum systems.

Story Source:
Materials provided by DOE/Argonne National Laboratory. Note: Content may be edited for style and length. More

Scientists at the University of California, Irvine have developed a new deep-learning framework that predicts gene regulation at the single-cell level.
Deep learning, a family of machine-learning methods based on artificial neural networks, has revolutionized applications such as image interpretation, natural language processing and autonomous driving. In a study published recently in Science Advances, UCI researchers describe how the technique can also be successfully used to observe gene regulation at the cellular level. Until now, that process had been limited to tissue-level analysis.
According to co-senior author Xiaohui Xie, UCI professor of computer science, the framework enables the study of transcription factor binding at the cellular level, which was previously impossible due to the intrinsic noise and sparsity of single-cell data. A transcription factor is a protein that controls the translation of genetic information from DNA to RNA; TFs regulate genes to ensure they’re expressed in proper sequence and at the right time in cells.
“The breakthrough was in realizing that we could leverage deep learning and massive datasets of tissue-level TF binding profiles to understand how TFs regulate target genes in individual cells through specific signals,” Xie said.
By training a neural network on large-scale genomic and epigenetic datasets, and by drawing on the expertise of collaborators across three departments, the researchers were able to identify novel gene regulations for individual cells or cell types.
“Our capability of predicting whether certain transcriptional factors are binding to DNA in a specific cell or cell type at a particular time provides a new way to tease out small populations of cells that could be critical to understanding and treating diseases,” said co-senior author Qing Nie, UCI Chancellor’s Professor of mathematics and director of the campus’s National Science Foundation-Simons Center for Multiscale Cell Fate Research, which supported the project.
He said that scientists can use the deep-learning framework to identify key signals in cancer stem cells — a small cell population that is difficult to specifically target in treatment or even quantify.
“This interdisciplinary project is a prime example of how researchers with different areas of expertise can work together to solve complex biological questions through machine-learning techniques,” Nie added.

Story Source:
Materials provided by University of California – Irvine. Note: Content may be edited for style and length. More

Researchers demonstrate a new technique for suppressing back reflections of light, leading to better signal quality for sensing and information technology.
Microresonators are small glass structures in which light can circulate and build up in intensity. Due to material imperfections, some amount of light is reflected backwards, which is disturbing their function.
Researchers have now demonstrated a method for suppressing these unwanted back reflections. Their findings can help improve a multitude of microresonator-based applications from measurement technology such as sensors used for example in drones, to optical information processing in fibre networks and computers.
The results of the team spanning the Max Planck Institute for the Science of Light (Germany), Imperial College London, and the National Physical Laboratory (UK) were recently published today in the Nature-family journal Light: Science and Applications.
Researchers and engineers are discovering many uses and applications for optical microresonators, a type of device often referred to as a light trap. One limitation of these devices is that they have some amount of back reflection, or backscattering, of light due to material and surface imperfections. The back reflected light negatively impacts of the usefulness of the tiny glass structures. To reduce the unwanted backscattering, the British and German scientists were inspired by noise cancelling headphones, but rather using optical than acoustic interference.
“In these headphones, out-of-phase sound is played to cancel out undesirable background noise,” says lead author Andreas Svela from the Quantum Measurement Lab at Imperial College London. “In our case, we are introducing out-of-phase light to cancel out the back reflected light,” Svela continues.
To generate the out-of-phase light, the researchers position a sharp metal tip close to the microresonator surface. Just like the intrinsic imperfections, the tip also causes light to scatter backwards, but there is an important difference: The phase of the reflected light can be chosen by controlling the position of the tip. With this control, the added backscattered light’s phase can be set so it annihilates the intrinsic back reflected light — the researchers produce darkness from light.
“It is an unintuitive result, by introducing an additional scatterer, we can reduce the total backscattering,” says co-author and principal investigator Pascal Del’Haye at the Max Planck Institute for the Science of Light. The published paper shows a record suppression of more than 30 decibels compared to the intrinsic back reflections. In other words, the unwanted light is less than a thousandth of what it was prior to applying the method.
“These findings are exciting as the technique can be applied to a wide range of existing and future microresonator technologies,” comments principal investigator Michael Vanner from the Quantum Measurement Lab at Imperial College London.
For example, the method can be used to improve gyroscopes, sensors that for instance help drones navigate; or to improve portable optical spectroscopy systems, opening for scenarios like built-in sensors in smartphones for detection of dangerous gasses or helping check the quality of groceries. Furthermore, optical components and networks with better signal quality allows us to transport more information even faster.

Story Source:
Materials provided by Imperial College London. Note: Content may be edited for style and length. More