More stories

  • in

    Research reveals rare metal could offer revolutionary switch for future quantum devices

    Quantum scientists have discovered a rare phenomenon that could hold the key to creating a ‘perfect switch’ in quantum devices which flips between being an insulator and superconductor.
    The research, led by the University of Bristol and published in Science, found these two opposing electronic states exist within purple bronze, a unique one-dimensional metal composed of individual conducting chains of atoms.
    Tiny changes in the material, for instance prompted by a small stimulus like heat or light, may trigger an instant transition from an insulating state with zero conductivity to a superconductor with unlimited conductivity, and vice versa. This polarised versatility, known as ’emergent symmetry’, has the potential to offer an ideal On/Off switch in future quantum technology developments.
    Lead author Nigel Hussey, Professor of Physics at the University of Bristol, said: “It’s a really exciting discovery which could provide a perfect switch for quantum devices of tomorrow.
    “The remarkable journey started 13 years ago in my lab when two PhD students, Xiaofeng Xu and Nick Wakeham, measured the magnetoresistance — the change in resistance caused by a magnetic field — of purple bronze.”
    In the absence of a magnetic field, the resistance of purple bronze was highly dependent on the direction in which the electrical current is introduced. Its temperature dependence was also rather complicated. Around room temperature, the resistance is metallic, but as the temperature is lowered, this reverses and the material appears to be turning into an insulator. Then, at the lowest temperatures, the resistance plummets again as it transitions into a superconductor. Despite this complexity, surprisingly, the magnetoresistance was found to be extremely simple. It was essentially the same irrespective of the direction in which the current or field were aligned and followed a perfect linear temperature dependence all the way from room temperature down to the superconducting transition temperature.
    “Finding no coherent explanation for this puzzling behaviour, the data lay dormant and published unpublished for the next seven years. A hiatus like this is unusual in quantum research, though the reason for it was not a lack of statistics,” Prof Hussey explained.

    “Such simplicity in the magnetic response invariably belies a complex origin and as it turns out, its possible resolution would only come about through a chance encounter.”
    In 2017, Prof Hussey was working at Radboud University and saw advertised a seminar by physicist Dr Piotr Chudzinski on the subject of purple bronze. At the time few researchers were devoting an entire seminar to this little-known material, so his interest was piqued.
    Prof Hussey said: “In the seminar Chudzinski proposed that the resistive upturn may be caused by interference between the conduction electrons and elusive, composite particles known as ‘dark excitons’. We chatted after the seminar and together proposed an experiment to test his theory. Our subsequent measurements essentially confirmed it.”
    Buoyed by this success, Prof Hussey resurrected Xu and Wakeham’s magnetoresistance data and showed them to Dr Chudzinski. The two central features of the data — the linearity with temperature and the independence on the orientation of current and field — intrigued Chudzinski, as did the fact that the material itself could exhibit both insulating and superconducting behaviour depending on how the material was grown.
    Dr Chudzinski wondered whether rather than transforming completely into an insulator, the interaction between the charge carriers and the excitons he’d introduced earlier could cause the former to gravitate towards the boundary between the insulating and superconducting states as the temperature is lowered. At the boundary itself, the probability of the system being an insulator or a superconductor is essentially the same.
    Prof Hussey said: “Such physical symmetry is an unusual state of affairs and to develop such symmetry in a metal as the temperature is lowered, hence the term ’emergent symmetry’, would constitute a world-first.”
    Physicists are well versed in the phenomenon of symmetry breaking: lowering the symmetry of an electron system upon cooling. The complex arrangement of water molecules in an ice crystal is an example of such broken symmetry. But the converse is an extremely rare, if not unique, occurrence. Returning to the water/ice analogy, it is as though upon cooling the ice further, the complexity of the ice crystals ‘melts’ once again into something as symmetric and smooth as the water droplet.

    Dr Chudzinski, now a Research Fellow at Queen’s University Belfast, said: “Imagine a magic trick where a dull, distorted figure transforms into a beautiful, perfectly symmetric sphere. This is, in a nutshell, the essence of emergent symmetry. The figure in question is our material, purple bronze, while our magician is nature itself.”
    To further test whether the theory held water, an additional 100 individual crystals, some insulating and others superconducting, were investigated by another PhD student, Maarten Berben, working at Radboud University.
    Prof Hussey added: “After Maarten’s Herculean effort, the story was complete and the reason why different crystals exhibited such wildly different ground states became apparent. Looking ahead, it might be possible to exploit this ‘edginess’ to create switches in quantum circuits whereby tiny stimuli induce profound, orders-of-magnitude changes in the switch resistance.” More

  • in

    Nostalgia and memories after ten years of social media

    As possibilities have changed and technology has advanced, memories and nostalgia are now a significant part of our use of social media. This is shown in a study from the University of Gothenburg and University West.
    Researchers at the University of Gothenburg and University West have been following a group of eleven active social media users for ten years, allowing them to describe and reflect on how they use the platforms to document and share their lives. The study provides insight into the role of technology in creating experiences and reliving meaningful moments.
    “These types of studies help us look back and understand the culture as it was in the 2010s and 2020s when social media was a central part of it,” says Beata Jungselius, senior lecturer of informatics at University West and one of the researchers behind the study.
    Social media users engage in what researchers define as “social media nostalgizing,” meaning they actively seek out content that evokes feelings of nostalgia.
    Alexandra Weilenmann, professor of interaction design at the University of Gothenburg, explains that participants in the study have described it as “treating themselves” to a nostalgia trip now and then.
    “Going back and remembering what has happened earlier in life becomes a bigger part of it over time than posting new content,” she says, and explains that in later interviews, it becomes clear that the platforms often serve as diary-like tools that allow memories to be relived.
    Social media platforms are introducing increasingly advanced features to help users interact with older content. Personal, music-infused photo albums generated for us or reminders of pictures we posted on the same date one, three, or ten years ago allow for nostalgic experiences, which are often seen as positive. The study describes how these features can lead to users reconnecting with old friends by “tagging” them in a shared memory. Alexandra Weilenmann and Beata Jungselius believe this could be a deliberate move by social media platforms to encourage users to stay active since the publication of new content has decreased.
    The researchers have noted that it’s not just the content itself that evokes feelings of nostalgia but also memories of the actual usage of social media play a significant role. For example, one of the interviewees reminisces about how rewarding the intense communication in forums was and how it often led to real-life meetings and interactions.
    “It’s only now that we’ve lived with social media long enough to make and draw conclusions from a study like this. Through our method of studying the same users over ten years, we’ve been able to follow how their usage and attitudes toward the platforms have changed as they have evolved,” says Beata Jungselius. More

  • in

    New computer code for mechanics of tissues and cells in three dimensions

    Biological materials are made of individual components, including tiny motors that convert fuel into motion. This creates patterns of movement, and the material shapes itself with coherent flows by constant consumption of energy. Such continuously driven materials are called “active matter.” The mechanics of cells and tissues can be described by active matter theory, a scientific framework to understand shape, flows, and form of living materials. The active matter theory consists of many challenging mathematical equations.
    Scientists from the Max Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG) in Dresden, the Center for Systems Biology Dresden (CSBD), and the TU Dresden have now developed an algorithm, implemented in an open-source supercomputer code, that can for the first time solve the equations of active matter theory in realistic scenarios. These solutions bring us a big step closer to solving the century-old riddle of how cells and tissues attain their shape and to designing artificial biological machines.
    Biological processes and behaviors are often very complex. Physical theories provide a precise and quantitative framework for understanding them. The active matter theory offers a framework to understand and describe the behavior of active matter — materials composed of individual components capable of converting a chemical fuel (“food”) into mechanical forces. Several scientists from Dresden were key in developing this theory, among others Frank Jülicher, director at the Max Planck Institute for the Physics of Complex Systems, and Stephan Grill, director at the MPI-CBG. With these principles of physics, the dynamics of active living matter can be described and predicted by mathematical equations. However, these equations are extremely complex and hard to solve. Therefore, scientists require the power of supercomputers to comprehend and analyze living materials. There are different ways to predict the behavior of active matter, with some focusing on the tiny individual particles, others studying active matter at the molecular level, and yet others studying active fluids on a large scale. These studies help scientists see how active matter behaves at different scales in space and over time.
    Solving complex mathematical equations
    Scientists from the research group of Ivo Sbalzarini, TU Dresden Professor at the Center for Systems Biology Dresden (CSBD), research group leader at the Max Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG), and Dean of the Faculty of Computer Science at TU Dresden, have now developed a computer algorithm to solve the equations of active matter. Their work was published in the journal “Physics of Fluids” and was featured on the cover. They present an algorithm that can solve the complex equations of active matter in three dimensions and in complex-shaped spaces. “Our approach can handle different shapes in three dimensions over time,” says one of the first authors of the study, Abhinav Singh, a studied mathematician. He continues, “Even when the data points are not regularly distributed, our algorithm employs a novel numerical approach that works seamlessly for complex biologically realistic scenarios to accurately solve the theory’s equations. Using our approach, we can finally understand the long-term behavior of active materials in both moving and non-moving scenarios for predicting their dynamics. Further, the theory and simulations could be used to program biological materials or create engines at the nano-scale to extract useful work.” The other first author, Philipp Suhrcke, a graduate of TU Dresden’s Computational Modeling and Simulation M.Sc. program, adds, “thanks to our work, scientists can now, for example, predict the shape of a tissue or when a biological material is going to become unstable or dysregulated, with far-reaching implications in understanding the mechanisms of growth and disease.”
    A powerful code for everyone to use
    The scientists implemented their software using the open-source library OpenFPM, meaning that it is freely available for others to use. OpenFPM is developed by the Sbalzarini group for democratizing large-scale scientific computing. The authors first developed a custom computer language that allows computational scientists to write supercomputer codes by specifying the equations in mathematical notation and let the computer do the work to create a correct program code. As a result, they do not have to start from scratch every time they write a code, effectively reducing code development times in scientific research from months or years to days or weeks, providing enormous productivity gains. Due to the tremendous computational demands of studying three-dimensional active materials, the new code is scalable on shared and distributed-memory multi-processor parallel supercomputers, thanks to the use of OpenFPM. Although the application is designed to run on powerful supercomputers, it can also run on regular office computers for studying two-dimensional materials.

    The Principal Investigator of the study, Ivo Sbalzarini, summarizes: “Ten years of our research went into creating this simulation framework and enhancing the productivity of computational science. This now all comes together in a tool for understanding the three-dimensional behavior of living materials. Open-source, scalable, and capable of handling complex scenarios, our code opens new avenues for modeling active materials. This may finally lead us to understand how cells and tissues attain their shape, addressing the fundamental question of morphogenesis that has puzzled scientist for centuries. But it may also help us design artificial biological machines with minimal numbers of components.”
    The computer code that support the findings of this study are openly available in the 3Dactive-hydrodynamics github repository located at https://github.com/mosaic-group/3Dactive-hydrodynamics
    The open source framework OpenFPM is available at https://github.com/mosaic-group/openfpm_pdata
    Related Publications for the embedded computer language and the OpenFPM software library: https://doi.org/10.1016/j.cpc.2019.03.007 and https://doi.org/10.1140/epje/s10189-021-00121-x More

  • in

    Breakthrough in tackling increasing demand by ‘internet of things’ on mobile networks

    A novel technology to manage demands on mobile networks from multiple users using Terahertz frequencies has been developed by University of Leicester computer scientists.
    As we see an explosion of devices joining the ‘internet of things’, this solution could not only improve speed and power consumption for users of mobile devices, but could also help reap the benefits from the next generation of mobile technologies, 6G.
    They have detailed the technology in a new study in IEEE Transactions on Communications.
    Demands on the UK’s mobile telecommunications network are growing, with Mobile UK estimating that twenty-five million devices are connected to mobile networks, a number expected to rise to thirty billion by 2030. As the ‘internet of things’ grows, more and more technology will be competing for access to those networks.
    State-of-the-art telecommunication technologies have been established for current applications in 5G, but with increasing demands of more users and devices, these systems demonstrate slower connections and costly energy consumption. These systems suffer from the self-interference problem that severely affects communication quality and efficiency. To deal with these challenges, a technique known as multicarrier-division duplex (MDD) has been recently proposed and studied, which allows a receiver in the network to be nearly free of self-interference in the digital domain by relying only on the fast Fourier transform (FFT) processing.
    This project proposed a novel technology to optimise the assignment of subcarrier set and the number of access point clusters and improve the communication quality in different networks. The team tested their technology in a simulation based on a real-world industrial setting, finding that it out-performed existing technologies. A 10% power consumption reduction can be achieved, compared to other state of the art technologies.
    Lead Principal Investigator Professor Huiyu Zhou from the University of Leicester School of Computing and Mathematical Sciences said: “With our proposed technology, 5G/6G systems require less energy consumption, have faster device selection and less resource allocation. Users may feel their mobile communication is quicker, wider and with reduced power demands.
    “The University of Leicester is leading the development of AI solutions for device selection and access point clustering. AI technologies, reinforcement learning in particular, help us to search for the best parameters used in the proposed wireless communication systems quickly and effectively. This helps to save power, resources and human labour. Without using AI technologies, we will spend much more time on rendering the best parameters for system set-up and device selection in the network.”
    The team is now continuing work on the optimising the proposed technologies and reducing the computational complexity of the technique. The source code of the proposed method has been published and shared with the entire world for promoting the research.
    The study forms part of the EU-funded 6G BRAINS project, which will develop an AI-driven self-learning platform to intelligently and dynamically allocate resources, enhancing capacity and reliability, and improving positioning accuracy while decreasing latency of response for future industrial applications of massive scale and varying demands. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101017226. More

  • in

    Shedding light on unique conduction mechanisms in a new type of perovskite oxide

    The remarkable proton and oxide-ion (dual-ion) conductivities of hexagonal perovskite-related oxide Ba7Nb3.8Mo1.2O20.1 are promising for next-generation electrochemical devices, as reported by scientists at Tokyo Tech. The unique ion-transport mechanisms they unveiled will hopefully pave the way for better dual-ion conductors, which could play an essential role in tomorrow’s clean energy technologies.
    Clean energy technologies are the cornerstone of sustainable societies, and solid-oxide fuel cells (SOFCs) and proton ceramic fuel cells (PCFCs) are among the most promising types of electrochemical devices for green power generation. These devices, however, still face challenges that hinder their development and adoption.
    Ideally, SOFCs should be operated at low temperatures to prevent unwanted chemical reactions from degrading their constituent materials. Unfortunately, most known oxide-ion conductors, a key component of SOFCs, only exhibit decent ionic conductivity at elevated temperatures. As for PCFCs, not only are they chemically unstable under carbon dioxide atmospheres, but they also require energy-intensive, high-temperature processing steps during manufacture.
    Fortunately, there is a type of material that can solve these problems by combining the benefits of both SOFCs and PCFCs: dual-ion conductors. By supporting the diffusion of both protons and oxide ions, dual-ion conductors can realize high total conductivity at lower temperatures and improve the performance of electrochemical devices. Although some perovskite-related dual-ion conducting materials such as Ba7Nb4MoO20 have been reported, their conductivities are not high enough for practical applications, and their underlying conducting mechanisms are not well understood.
    Against this backdrop, a research team led by Professor Masatomo Yashima from Tokyo Institute of Technology, Japan, decided to investigate the conductivity of materials similar to 7Nb4MoO20 but with a higher Mo fraction (that is, Ba7Nb4-xMo1+xO20+x/2). Their latest study, which was conducted in collaboration with the Australian Nuclear Science and Technology Organisation (ANSTO), the High Energy Accelerator Research Organization (KEK), and Tohoku University, was published in Chemistry of Materials.
    After screening various Ba7Nb4-xMo1+xO20+x/2 compositions, the team found that Ba7Nb3.8Mo1.2O20.1 had remarkable proton and oxide-ion conductivities. “Ba7Nb3.8Mo1.2O20.1 exhibited bulk conductivities of 11 mS/cm at 537 ℃ under wet air and 10 mS/cm at 593 ℃ under dry air. Total direct current conductivity at 400 ℃ in wet air of Ba7Nb3.8Mo1.2O20.1 was 13 times higher than that of Ba7Nb4MoO20, and the bulk conductivity in dry air at 306 ℃ is 175 times higher than that of the conventional yttria-stabilized zirconia (YSZ),” highlights Prof. Yashima.
    Next, the researchers sought to shed light on the underlying mechanisms behind these high conductivity values. To this end, they conducted ab initio molecular dynamics (AIMD) simulations, neutron diffraction experiments, and neutron scattering length density analyses. These techniques enabled them to study the structure of Ba7Nb3.8Mo1.2O20.1 in greater detail and determine what makes it special as a dual-ion conductor.
    Interestingly, the team found that the high oxide-ion conductivity of Ba7Nb3.8Mo1.2O20.1 originates from a unique phenomenon. It turns out that adjacent MO5 monomers in Ba7Nb3.8Mo1.2O20.1 can form M2O9 dimers by sharing an oxygen atom on one of their corners (M = Nb or Mo cation). The breaking and reforming of these dimers gives rise to ultrafast oxide-ion movement in a manner analogous to a long line of people relaying buckets of water (oxide ions) from one person to the next. Furthermore, the AIMD simulations revealed that the observed high proton conduction was due to efficient proton migration in the hexagonal close-packed BaO3 layers in the material.
    Taken together, the results of this study highlight the potential of perovskite-related dual-ion conductors and could serve as guidelines for the rational design of these materials. “The present findings of high conductivities and unique ion migration mechanisms in Ba7Nb3.8Mo1.2O20.1 will help the development of science and engineering of oxide-ion, proton, and dual-ion conductors,” concludes a hopeful Prof. Yashima. More

  • in

    Future of brain-inspired AI as Python code library passes major milestone

    Four years ago, UC Santa Cruz’s Jason Eshraghian developed a Python library that combines neuroscience with artificial intelligence to create spiking neural networks, a machine learning method that takes inspiration from the brain’s ability to efficiently process data. Now, his open source code library, called “snnTorch,” has surpassed 100,000 downloads and is used in a wide variety of projects, from NASA satellite tracking efforts to semiconductor companies optimizing chips for AI.
    A new paper published in the journal Proceedings of the IEEE documents the coding library but also is intended to be a candid educational resource for students and any other programmers interested in learning about brain-inspired AI.
    “It’s exciting because it shows people are interested in the brain, and that people have identified that neural networks are really inefficient compared to the brain,” said Eshraghian, an assistant professor of electrical and computer engineering. “People are concerned about the environmental impact [of the costly power demands] of neural networks and large language models, and so this is a very plausible direction forward.”
    Building snnTorch
    Spiking neural networks emulate the brain and biological systems to process information more efficiently. The brain’s neurons are at rest until there is a piece of information for them to process, which causes their activity to spike. Similarly, a spiking neural network only begins processing data when there is an input into the system, rather than constantly processing data like traditional neural networks.
    “We want to take all the benefits of the brain and its power efficiency and smush them into the functionality of artificial intelligence — so taking the best of both worlds,” Eshraghian said.
    Eshraghian began building the code for a spiking neural network in Python as a passion project during the pandemic, somewhat as a method to teach himself the coding language Python. A chip designer by training, he became interested in learning to code when considering that computing chips could be optimized for power efficiency by co-designing the software and the hardware to ensure they best complement each other.

    Now, snnTorch is being used by thousands of programmers around the world on a variety of projects, supporting everything from NASA’s satellite tracking projects to major chip designers such as Graphcore.
    While building the Python library, Eshraghian created code documentation and educational materials, which came naturally to him in the process of teaching himself the coding language. The documents, tutorials, and interactive coding notebooks he made later exploded in the community and became the first point of entry for many people learning about the topics of neuromorphic engineering and spiking neural networks, which he sees as one of the major reasons that his library became so popular.
    An honest resource
    Knowing that these educational materials could be very valuable to the growing community of computer scientists and beyond who were interested in the field, Eshraghian began compiling his extensive documentation into a paper, which has now been published in the Proceedings of the IEEE, a leading computing journal.
    The paper acts as a companion to the snnTorch code library and is structured like a tutorial, and an opinionated one at that, discussing uncertainty among brain-inspired deep learning researchers and offering a perspective on the future of the field. Eshraghian said that the paper is intentionally upfront to its readers that the field of neuromorphic computing is evolving and unsettled in an effort to save students the frustration of trying to find the theoretical basis for code decision-making that the research community doesn’t even understand.
    “This paper is painfully honest, because students deserve that,” Eshraghian said. “There’s a lot of things that we do in deep learning, and we just don’t know why they work. A lot of times we want to claim that we did something intentionally, and we published because we went through a series of rigorous experiments, but here we say just: this is what works best and we have no idea why.”
    The paper contains blocks of code, a format unusual to typical research papers. These code blocks are sometimes accompanied by explanations that certain areas may be vastly unsettled, but provide insight into why researchers think certain approaches may be successful. Eshraghian said he has seen a positive reception to this honest approach in the community, and has even been told that the paper is being used in onboarding materials at neuromorphic hardware startups.

    “I don’t want my research to put people through the same pain I went through,” he said.
    Learning from and about the brain
    The paper offers a perspective on how researchers in the field might navigate some of the limitations of brain-inspired deep learning that stem from the fact that overall, our understanding of how the brain functions and processes information is quite limited.
    For AI researchers to move toward more brain-like learning mechanisms for their deep learning models, they need to identify the correlations and discrepancies between deep learning and biology, Eshraghian said. One of these key differences is that brains can’t survey all of the data they’ve ever inputted in the way that AI models can, and instead focus on the real-time data that comes their way, which could offer opportunities for enhanced energy efficiency.
    “Brains aren’t time machines, they can’t go back — all your memories are pushed forward as you experience the world, so training and processing are coupled together,” Eshraghian said. “One of the things that I make a big deal of in the paper is how we can apply learning in real time.”
    Another area of exploration in the paper is a fundamental concept in neuroscience that states that neurons that fire together are wired together — meaning when two neurons are triggered to send out a signal at the same time, the pathway between the two neurons is strengthened. However, the ways in which the brain learns on an organ-wide scale still remains mysterious.
    The “fire together, wired together” concept has been traditionally seen as in opposition to deep learning’s model training method known as backpropagation, but Eshraghian suggests that these processes may be complementary, opening up new areas of exploration for the field.
    Eshraghian is also excited about working with cerebral organoids, which are models of brain tissue grown from stem cells, to learn more about how the brain processes information. He’s currently collaborating with biomolecular engineering researchers in the UCSC Genomics Institute’s Braingeneers group to explore these questions with organoid models. This is a unique opportunity for UC Santa Cruz engineers to incorporate “wetware” — a term referring to biological models for computing research — into the software/hardware co-design paradigm that is prevalent in the field. The snnTorch code could even provide a platform for simulating organoids, which can be difficult to maintain in the lab.
    “[The Braingeneers] are building the biological instruments and tools that we can use to get a better feel for how learning can happen, and how that might translate in order to make deep learning more efficient,” Eshraghian said.
    Brain-inspired learning at UCSC and beyond
    Eshraghian is now using the concepts developed in his library and the recent paper in his class on neuromorphic computing at UC Santa Cruz called “Brain-Inspired Deep Learning.” Undergraduate and graduate students across a range of academic disciplines are taking the class to learn the basics of deep learning and complete a project in which they write their own tutorial for, and potentially contributing to, snnTorch.
    “It’s not just kind of coming out of the class with an exam or getting an A plus, it’s now making a contribution to something, and being able to say that you’ve done something tangible,” Eshraghian said.
    Meanwhile, the preprint version of the recent IEEE paper continues to receive contributions from researchers around the world, a reflection of the dynamic, open-source nature of the field. A new NSF grant he is a co-principal investigator on will support students’ ability to attend the month-long Telluride Neuromorphic & Cognition Engineering workshop.
    Eshraghian is collaborating with people to push the field in a number of ways, from making biological discoveries about the brain, to pushing the limits of neuromorphic chips to handle low-power AI workloads, to facilitating collaboration to bring the spiking neural network-style of computing to other domains such as natural physics.
    Discord and Slack channels dedicated to discussing the spiking neural network code support a thriving environment of collaboration across industry and academia. Eshraghian even recently came across a job posting that listed proficiency in snnTorch as a desired quality. More

  • in

    Dams now run smarter with AI

    In August 2020, following a period of prolonged drought and intense rainfall, a dam situated near the Seomjin River in Korea experienced overflow during a water release, resulting in damages exceeding 100 billion won (USD 76 million). The flooding was attributed to maintaining the dam’s water level 6 meters higher than the norm. Could this incident have been averted through predictive dam management?
    A research team led by Professor Jonghun Kam and Eunmi Lee, a PhD candidate, from the Division of Environmental Science & Engineering at Pohang University of Science and Technology (POSTECH), recently employed deep learning techniques to scrutinize dam operation patterns and assess their effectiveness. Their findings were published in the Journal of Hydrology.
    Korea faces a precipitation peak during the summer, relying on dams and associated infrastructure for water management. However, the escalating global climate crisis has led to the emergence of unforeseen typhoons and droughts, complicating dam operations. In response, a new study has emerged, aiming to surpass conventional physical models by harnessing the potential of an artificial intelligence (AI) model trained on extensive big data.
    The team focused on crafting an AI model aimed at not only predicting the operational patterns of dams within the Seomjin River basin, specifically focusing on the Seomjin River Dam, Juam Dam, and Juam Control Dam, but also understanding the decision-making processes of the trained AI models. Their objective was to formulate a scenario outlining the methodology for forecasting dam water levels. Employing the Gated Recurrent Unit (GRU) model, a deep learning algorithm, the team trained it using data spanning from 2002 to 2021 from dams along the Seomjin River. Precipitation, inflow, and outflow data served as inputs while hourly dam levels served as outputs. The analysis demonstrated remarkable accuracy, boasting an efficiency index exceeding 0.9.
    Subsequently, the team devised explainable scenarios, manipulating inputs by -40%, -20%, +20%, and 40%, of each input variable to examine how the trained GRU model responded to these alterations in inputs. While changes in precipitation had a negligible impact on dam water levels, variations in inflow significantly influenced the dam’s water level. Notably, the identical change in outflow yielded different water levels at distinct dams, affirming that the GRU model had effectively learned the unique operational nuances of each dam.
    Professor Jonghun Kam remarked “Our examination delved beyond predicting the patterns of dam operations securitize their effectiveness using AI models. We introduced a methodology aimed at indirectly understanding the decision-making process of AI-based black box model determining dam water levels.” He further stated, “Our aspiration is that this insight will contribute to a deeper understanding of dam operations and enhance their efficiency in the future.”
    The research was sponsored by the Mid-career Researcher Program of the National Research Foundation of Korea. More

  • in

    The mind’s eye of a neural network system

    In the background of image recognition software that can ID our friends on social media and wildflowers in our yard are neural networks, a type of artificial intelligence inspired by how own our brains process data. While neural networks sprint through data, their architecture makes it difficult to trace the origin of errors that are obvious to humans — like confusing a Converse high-top with an ankle boot — limiting their use in more vital work like health care image analysis or research. A new tool developed at Purdue University makes finding those errors as simple as spotting mountaintops from an airplane.
    “In a sense, if a neural network were able to speak, we’re showing you what it would be trying to say,” said David Gleich, a Purdue professor of computer science in the College of Science who developed the tool, which is featured in a paper published in Nature Machine Intelligence. “The tool we’ve developed helps you find places where the network is saying, ‘Hey, I need more information to do what you’ve asked.’ I would advise people to use this tool on any high-stakes neural network decision scenarios or image prediction task.”
    Code for the tool is available on GitHub, as are use case demonstrations. Gleich collaborated on the research with Tamal K. Dey, also a Purdue professor of computer science, and Meng Liu, a former Purdue graduate student who earned a doctorate in computer science.
    In testing their approach, Gleich’s team caught neural networks mistaking the identity of images in databases of everything from chest X-rays and gene sequences to apparel. In one example, a neural network repeatedly mislabeled images of cars from the Imagenette database as cassette players. The reason? The pictures were drawn from online sales listings and included tags for the cars’ stereo equipment.
    Neural network image recognition systems are essentially algorithms that process data in a way that mimics the weighted firing pattern of neurons as an image is analyzed and identified. A system is trained to its task — such as identifying an animal, a garment or a tumor — with a “training set” of images that includes data on each pixel, tagging and other information, and the identity of the image as classified within a particular category. Using the training set, the network learns, or “extracts,” the information it needs in order to match the input values with the category. This information, a string of numbers called an embedded vector, is used to calculate the probability that the image belongs to each of the possible categories. Generally speaking, the correct identity of the image is within the category with the highest probability.
    But the embedded vectors and probabilities don’t correlate to a decision-making process that humans would recognize. Feed in 100,000 numbers representing the known data, and the network produces an embedded vector of 128 numbers that don’t correspond to physical features, although they do make it possible for the network to classify the image. In other words, you can’t open the hood on the algorithms of a trained system and follow along. Between the input values and the predicted identity of the image is a proverbial “black box” of unrecognizable numbers across multiple layers.
    “The problem with neural networks is that we can’t see inside the machine to understand how it’s making decisions, so how can we know if a neural network is making a characteristic mistake?” Gleich said.

    Rather than trying to trace the decision-making path of any single image through the network, Gleich’s approach makes it possible to visualize the relationship that the computer sees among all the images in an entire database. Think of it like a bird’s-eye view of all the images as the neural network has organized them.
    The relationship among the images (like network’s prediction of the identity classification of each of the images in the database) is based on the embedded vectors and probabilities the network generates. To boost the resolution of the view and find places where the network can’t distinguish between two different classifications, Gleich’s team first developed a method of splitting and overlapping the classifications to identify where images have a high probability of belonging to more than one classification.
    The team then maps the relationships onto a Reeb graph, a tool taken from the field of topological data analysis. On the graph, each group of images the network thinks are related is represented by a single dot. Dots are color coded by classification. The closer the dots, the more similar the network considers groups to be, and most areas of the graph show clusters of dots in a single color. But groups of images with a high probability of belonging to more than one classification will be represented by two differently colored overlapping dots. With a single glance, areas where the network cannot distinguish between two classifications appear as a cluster of dots in one color, accompanied by a smattering of overlapping dots in a second color. Zooming in on the overlapping dots will show an area of confusion, like the picture of the car that’s been labeled both car and cassette player.
    “What we’re doing is taking these complicated sets of information coming out of the network and giving people an ‘in’ into how the network sees the data at a macroscopic level,” Gleich said. “The Reeb map represents the important things, the big groups and how they relate to each other, and that makes it possible to see the errors.”
    “Topological Structure of Complex Predictions” was produced with the support of the National Science Foundation and the U.S. Department of Energy. More