Computers Math

Artificial intelligence tools and deep learning models are a powerful tool in cancer treatment. They can be used to analyze digital images of tumor biopsy samples, helping physicians quickly classify the type of cancer, predict prognosis and guide a course of treatment for the patient. However, unless these algorithms are properly calibrated, they can sometimes make inaccurate or biased predictions.
A new study led by researchers from the University of Chicago shows that deep learning models trained on large sets of cancer genetic and tissue histology data can easily identify the institution that submitted the images. The models, which use machine learning methods to “teach” themselves how to recognize certain cancer signatures, end up using the submitting site as a shortcut to predicting outcomes for the patient, lumping them together with other patients from the same location instead of relying on the biology of individual patients. This in turn may lead to bias and missed opportunities for treatment in patients from racial or ethnic minority groups who may be more likely to be represented in certain medical centers and already struggle with access to care.
“We identified a glaring hole in the in the current methodology for deep learning model development which makes certain regions and patient populations more susceptible to be included in inaccurate algorithmic predictions,” said Alexander Pearson, MD, PhD, assistant Assistant Professor of Medicine at UChicago Medicine and co-senior author. The study was published July 20, in Nature Communications.
One of the first steps in treatment for a cancer patient is taking a biopsy, or small tissue sample of a tumor. A very thin slice of the tumor is affixed to glass slide, which is stained with multicolored dyes for review by a pathologist to make a diagnosis. Digital images can then be created for storage and remote analysis by using a scanning microscope. While these steps are mostly standard across pathology labs, minor variations in the color or amount of stain, tissue processing techniques and in the imaging equipment can create unique signatures, like tags, on each image. These location-specific signatures aren’t visible to the naked eye, but are easily detected by powerful deep learning algorithms.
These algorithms have the potential to be a valuable tool for allowing physicians to quickly analyze a tumor and guide treatment options, but the introduction of this kind of bias means that the models aren’t always basing their analysis on the biological signatures it sees in the images, but rather the image artifacts generated by differences between submitting sites.
Pearson and his colleagues studied the performance of deep learning models trained on data from the Cancer Genome Atlas, one of the largest repositories of cancer genetic and tissue image data. These models can predict survival rates, gene expression patterns, mutations, and more from the tissue histology, but the frequency of these patient characteristics varies widely depending on which institutions submitted the images, and the model often defaults to the “easiest” way to distinguish between samples — in this case, the submitting site.
For example, if Hospital A serves mostly affluent patients with more resources and better access to care, the images submitted from that hospital will generally indicate better patient outcomes and survival rates. If Hospital B serves a more disadvantaged population that struggles with access to quality care, the images that site submitted will generally predict worse outcomes.
The research team found that once the models identified which institution submitted the images, they tended to use that as a stand in for other characteristics of the image, including ancestry. In other words, if the staining or imaging techniques for a slide looked like it was submitted by Hospital A, the models would predict better outcomes, whereas they would predict worse outcomes if it looked like an image from Hospital B. Conversely, if all patients in Hospital B had biological characteristics based on genetics that indicated a worse prognosis, the algorithm would link the worse outcomes to Hospital B’s staining patterns instead of things it saw in the tissue.
“Algorithms are designed to find a signal to differentiate between images, and it does so lazily by identifying the site,” Pearson said. “We actually want to understand what biology within a tumor is more likely to predispose resistance to treatment or early metastatic disease, so we have to disentangle that site-specific digital histology signature from the true biological signal.”
The key to avoiding this kind of bias is to carefully consider the data used to train the models. Developers can make sure that different disease outcomes are distributed evenly across all sites used in the training data, or by isolating a certain site while training or testing the model when the distribution of outcomes is unequal. The result will produce more accurate tools that can get physicians the information they need to quickly diagnose and plan treatments for cancer patients.
“The promise of artificial intelligence is the ability to bring accurate and rapid precision health to more people,” Pearson said. “In order to meet the needs of the disenfranchised members of our society, however, we have to be able to develop algorithms which are competent and make relevant predictions for everyone.” More

It’s not exactly X-ray vision, but it’s close. In research published in the journal Optica, University of California, Irvine researchers describe a new type of camera technology that, when aimed at an object, can rapidly retrieve 3D images, displaying its chemical content down to the micrometer scale. The new tech promises to help companies inspect things like the insides of computer chips without having to pry them open — an advancement the researchers say could accelerate the production time of such goods by more than a hundred times.
“This is a paper about a way to visualize things in 3D very fast, even at video rate,” said Dmitry Fishman — director of laser spectroscopy labs in the UCI Department of Chemistry — who, along with Eric Potma, professor of chemistry, spearheaded the work. The novel imaging tech is based on a so-called nonlinear optical effect in silicon — a semiconductor material used in visible-light cameras and detectors.
Through such a nonlinear optical effect, conventional silicon detectors can sense light coming from the mid-infrared range of the electromagnetic spectrum. The reason being, Fishman explained, is that the mid-infrared spectral region carries important information on the material’s chemical make-up. “Most molecular vibrations and signatures are in the mid-infrared range,” he said.
Other technologies, he explained, are slow to retrieve images, because the laser light needs to scan across the object — a process that takes a longer amount of time. “A nonlinear optical ‘trick’ with short laser pulses allowed us to capture a depth-resolved image on a camera in one shot, thus providing an alternative method to what other people are doing — and the advance is that this is not just faster, but also produces 3D images with chemical contrast,” Fishman said.
And the imaging technology isn’t just for computer chips. Potma explained that the system can also image things like ceramics used to make things like heat shield plates on space shuttles and reveal clues about any structural weaknesses that might be there.
The research follows in the wake of work by Potma and Fishman and a team of researchers published last year in Nature’s Light: Science & Applications that describes the first steps toward creating efficient mid-infrared detection technology using off-the-shelf silicon-based cameras. Back then, the technology was just beginning to take shape, but now, Fishman explained, it’s getting close to being ready for the mainstream. “This time we made it much more efficient and better,” he said.
Funding for the work came from the National Institutes of Health and the National Science Foundation. The work was done in collaboration between UCI scientists and Yong Chen, a professor in the Epstein Department of Industrial & Systems Engineering at the University of Southern California.
Story Source:
Materials provided by University of California – Irvine. Note: Content may be edited for style and length. More

Many applications, from fiber-optic telecommunications to biomedical imaging processes require substances that emit light in the near-infrared range (NIR). A research team in Switzerland has now developed the first chromium complex that emits light in the coveted, longer wavelength NIR-II range. In the journal Angewandte Chemie, the team has introduced the underlying concept: a drastic change in the electronic structure of the chromium caused by the specially tailored ligands that envelop it.
Many materials that emit NIR light are based on expensive or rare metal complexes. Cheaper alternatives that emit in the NIR-I range between 700 and 950 nm have been developed but NIR-II-emitting complexes of non-precious metals remain extremely rare. Luminescence in the NIR-II range (1000 to 1700 nm) is, for example, particularly advantageous for in vivo imaging because this light penetrates very far into tissues.
The luminescence of complexes is based on the excitement of electrons, through the absorption of light, for example. When the excited electron drops back down to its ground state, part of the energy is emitted as radiation. The wavelength of this radiation depends on the energetic differences between the electronic states. In complexes, these are significantly determined by the type and arrangement of the ligands bound to the metal.
In typical chemical (covalent) bonds, each partner brings one electron to share in a bonding pair; in many complexes both of the electrons come from the ligand. However, the line between these types of bonds is fluid: metal-ligand bonds can have partial covalent character (nephelauxetic effect). As a consequence, the energy of certain excited states is reduced, giving the emitted radiation a longer wavelength. This has been observed for polypyridine ligands, which cause the ruby red emission of trivalent chromium (Cr(III)) in complexes to shift into the NIR-I range.
In order to increase the covalence of the metal-ligand bond and further increase the wavelength, Narayan Sinha in a team led by Claude Piguet and Oliver S. Wenger at the Universities of Basel and Geneva (Switzerland) switched from classic polypyridine ligands to a newly tailored, charged, tridentate chelate ligand. The term chelate is derived from the Greek word for the pincer of a crab, and tridentate means that the ligand has three binding sites with which it grabs the central metal ion like a pincer.
In the resulting new complex, the Cr(III) ion is surrounded on all sides by two tridentate charged chelate ligands to form an octahedral shape. This results in a drastically altered, unusual electronic structure with a high electron density on the Cr(III). In the axial direction, charge transfer takes place from the ligands to the metal, but in the equatorial plane of the octahedron, charge transfer moves from the metal to the ligands. The combined “push” and “pull” interactions likely have a strong influence on the spectroscopically relevant electrons of the Cr(III) — the key to the NIR-II emissions of the new complex.
Story Source:
Materials provided by Wiley. Note: Content may be edited for style and length. More

This week, at the 38th International Conference on Machine Learning (ICML 21), researchers at the NYU Center for Cyber Security at the NYU Tandon School of Engineering are revealing new insights into the basic functions that drive the ability of neural networks to make inferences on encrypted data.
In the paper, “DeepReDuce: ReLU Reduction for Fast Private Inference,” the team focuses on linear and non-linear operators, key features of neural network frameworks that, depending on the operation, introduce a heavy toll in time and computational resources. When neural networks compute on encrypted data, many of these costs are incurred by rectified linear activation function (ReLU), a non-linear operation.
Brandon Reagen, professor of computer science and engineering and electrical and computer engineering and a team of collaborators including Nandan Kumar Jha, a Ph.D. student, and Zahra Ghodsi, a former doctoral student under the guidance of Siddharth Garg, developed a framework called DeepReDuce. It offers a solution through rearrangement and reduction of ReLUs in neural networks.
Reagen explained that this shift requires a fundamental reassessment of where and how many components are distributed in neural networks systems.
“What we are trying to do is rethink how neural nets are designed in the first place,” he explained. “You can skip a lot of these time- and computationally-expensive ReLU operations and still get high performing networks at 2 to 4 times faster run time.”
The team found that, compared to the state-of-the-art for private inference, DeepReDuce improved accuracy and reduced ReLU count by up to 3.5% and 3.5×, respectively.
The inquiry is not merely academic. As the use of AI grows in concert with concerns about the security of personal, corporate, and government data security, neural networks are increasingly making computations on encrypted data. In such scenarios involving neural networks generating private inferences (PI’s) on hidden data without disclosing inputs, it is the non-linear functions that exert the highest “cost” in time and power. Because these costs increase the difficulty and time it takes for learning machines to do PI, researchers have struggled to lighten the load ReLUs exert on such computations.
The team’s work builds on innovative technology called CryptoNAS. Described in an earlier paper whose authors include Ghodsi and a third Ph.D. student, Akshaj Veldanda, CryptoNAS optimizes the use of ReLUs as one might rearrange how rocks are arranged in a stream to optimize the flow of water: it rebalances the distribution of ReLUS in the network and removes redundant ReLUs.
DeepReDuce expands on CryptoNAS by streamlining the process further. It comprises a set of optimizations for the judicious removal of ReLUs after CryptoNAS reorganization functions. The researchers tested DeepReDuce by using it to remove ReLUs from classic networks, finding that they were able to significantly reduce inference latency while maintaining high accuracy.
Reagan, with Mihalis Maniatakos, research assistant professor of electrical and computer engineering, is also part of a collaboration with data security company Duality to design a new microchip designed to handle computation on fully encrypted data.
The research on ReLUS was supported by ADA and the Data Protection in Virtual Environments (DPRIVE) program at the U.S. Defense Advanced Research Projects Agency (DARPA) and the Center for Applications Driving Architectures.
Story Source:
Materials provided by NYU Tandon School of Engineering. Note: Content may be edited for style and length. More

Exoskeletons — wearable devices used by workers on assembly lines or in warehouses to alleviate stress on their lower backs — may compete with valuable resources in the brain while people work, canceling out the physical benefits of wearing them, a new study suggests.
The study, published recently in the journal Applied Ergonomics, found that when people wore exoskeletons while performing tasks that required them to think about their actions, their brains worked overtime and their bodies competed with the exoskeletons rather than working in harmony with them. The study indicates that exoskeletons may place enough burden on the brain that potential benefits to the body are negated.
“It’s almost like dancing with a really bad partner,” said William Marras, senior author of the study, professor of integrated systems engineering and director of The Ohio State University Spine Research Institute.
“The exoskeleton is trying to anticipate your moves, but it’s not going well, so you fight with the exoskeleton, and that causes this change in your brain which changes the muscle recruitment — and could cause higher forces on your lower back, potentially leading to pain and possible injuries.”
For the study, researchers asked 12 people — six men and six women — to repeatedly lift a medicine ball in two 30-minutes sessions. For one of the sessions, the participants wore an exoskeleton. For the other, they did not.
The exoskeleton, which is attached to the user’s chest and legs, is designed to help control posture and motion during lifting to protect the lower back and reduce the possibility of injury. More

Researchers from the University of Southern California (USC) Department of Computer Science and NVIDIA have unveiled a new simulator for robotic cutting that can accurately reproduce the forces acting on a knife as it slices through common foodstuffs, such as fruit and vegetables. The system could also simulate cutting through human tissue, offering potential applications in surgical robotics. The paper was presented at the Robotics: Science and Systems (RSS) Conference 2021 on July 16, where it received the Best Student Paper Award.
In the past, researchers have had trouble creating intelligent robots that replicate cutting. One challenge: in the real world, no two objects are the same, and current robotic cutting systems struggle with variation. To overcome this, the team devised a unique approach to simulate cutting by introducing springs between the two halves of the object being cut, represented by a mesh. These springs are weakened over time in proportion to the force exerted by the knife on the mesh.
“What makes ours a special kind of simulator is that it is ‘differentiable,’ which means that it can help us automatically tune these simulation parameters from real-world measurements,” said lead author Eric Heiden, a PhD in computer science student at USC. “That’s important because closing this reality gap is a significant challenge for roboticists today. Without this, robots may never break out of simulation into the real world.”
To transfer skills from simulation to reality, the simulator must be able to model a real system. In one of the experiments, the researchers used a dataset of force profiles from a physical robot to produce highly accurate predictions of how the knife would move in real life. In addition to applications in the food processing industry, where robots could take over dangerous tasks like repetitive cutting, the simulator could improve force haptic feedback accuracy in surgical robots, helping to guide surgeons and prevent injury.
“Here, it is important to have an accurate model of the cutting process and to be able to realistically reproduce the forces acting on the cutting tool as different kinds of tissue are being cut,” said Heiden. “With our approach, we are able to automatically tune our simulator to match such different types of material and achieve highly accurate simulations of the force profile.” In ongoing research, the team is applying the system to real-world robots.
Co-authors are Miles Macklin, Yashraj S Narang, Dieter Fox, Animesh Garg, Fabio Ramos, all of NVIDIA.
Video: https://www.youtube.com/watch?v=bN4yqHhfAfQ
The full paper (open access) and blog post are available here: https://diff-cutting-sim.github.io/
Story Source:
Materials provided by University of Southern California. Original written by Caitlin Dawson. Note: Content may be edited for style and length. More

A team of scientists at Los Alamos National Laboratory proposes that modulated quantum metasurfaces can control all properties of photonic qubits, a breakthrough that could impact the fields of quantum information, communications, sensing and imaging, as well as energy and momentum harvesting. The results of their study were released yesterday in the journal Physical Review Letters, published by the American Physical Society.
“People have studied classical metasurfaces for a long time,” says Diego Dalvit, who works in the Condensed Matter and Complex Systems group at the Laboratory’s Theoretical Division. “But we came up with this new idea, which was to modulate in time and space the optical properties of a quantum metasurface that allow us to manipulate, on-demand, all degrees of freedom of a single photon, which is the most elementary unit of light.”
Metasurfaces are ultrathin structures that can manipulate light in ways not usually seen in nature. In this case, the team developed a metasurface that looked like an array of rotated crosses, which they can then manipulate with lasers or electrical pulses. They then proposed to shoot a single photon through the metasurface, where the photon splits into a superposition of many colors, paths, and spinning states that are all intertwined, generating so-called quantum entanglement — meaning the single photon is capable of inheriting all these different properties at once.
“When the metasurface is modulated with laser or electrical pulses, one can control the frequency of the refracted single photon, alter its angle of trajectory, the direction of its electric field, as well as its twist,” says Abul Azad from the Center for Integrated Nanotechnologies at the Laboratory’s Materials Physics and Applications Division.
By manipulating these properties, this technology could be used to encode information in photons traveling within a quantum network, everything from banks, quantum computers, and between Earth and satellites. Encoding photons is particularly desirable in the field of cryptography because “eavesdroppers” are unable to view a photon without changing its fundamental physics, which if done would then alert the sender and receiver that the information has been compromised.
The researchers are also working on how to pull photons from a vacuum by modulating the quantum metasurface.
“The quantum vacuum is not empty but full of fleeting virtual photons. With the modulated quantum metasurface one is able to efficiently extract and convert virtual photons into real photon pairs,” says Wilton Kort-Kamp, who works in the Theoretical Division at the Lab’s Condensed Matter and Complex Systems group.
Harnessing photons that exist in the vacuum and shooting them in one direction should create propulsion in the opposite direction. Similarly, stirring the vacuum should create rotational motion from the twisted photons. Structured quantum light could then one day be used to generate mechanical thrust, using only tiny amounts of energy to drive the metasurface.
Story Source:
Materials provided by DOE/Los Alamos National Laboratory. Note: Content may be edited for style and length. More

Northwestern University researchers have developed a new framework using machine learning that improves the accuracy of interatomic potentials — the guiding rules describing how atoms interact — in new materials design. The findings could lead to more accurate predictions of how new materials transfer heat, deform, and fail at the atomic scale.
Designing new nanomaterials is an important aspect of developing next-generation devices used in electronics, sensors, energy harvesting and storage, optical detectors, and structural materials. To design these materials, researchers create interatomic potentials through atomistic modeling, a computational approach that predicts how these materials behave by accounting for their properties at the smallest level. The process to establish materials’ interatomic potential — called parameterization — has required significant chemical and physical intuition, leading to less accurate prediction of new materials design.
The researchers’ platform minimizes user intervention by employing multi-objective genetic algorithm optimization and statistical analysis techniques, and screens promising interatomic potentials and parameter sets.
“The computational algorithms we developed provide analysts with a methodology to assess and avoid traditional shortcomings,” said Horacio Espinosa, James N. and Nancy J. Farley Professor in Manufacturing and Entrepreneurship and professor of mechanical engineering and (by courtesy) biomedical engineering and civil and environmental engineering, who led the research. “They also provide the means to tailor the parameterization to applications of interest.”
The findings were published in a study titled “Parametrization of Interatomic Potentials for Accurate Large Deformation Pathways Using Multi-Objective Genetic Algorithms and Statistical Analyses: A Case Study on Two-Dimensional Materials” on July 21 in Nature Partner Journals — Computational Materials.
Xu Zhang and Hoang Nguyen, both students in Northwestern Engineering’s Theoretical and Applied Mechanics (TAM) graduate program, were co-first authors of the study. Other co-authors included Jeffrey T. Paci of the University of Victoria, Canada, Subramanian Sankaranarayanan of Argonne National Laboratory, and Jose Mendoza of Michigan State University. More