Interferometric Synthetic Aperture Radar (InSAR) technology uses satellite radar to detect surface deformation patterns and monitor earthquake impacts on buildings. While vital for emergency response planning, extracting multi-class building damage classifications from InSAR data faces challenges: overlapping damage signatures with environmental noise, computational complexity in multi-class scenarios, and the need for rapid regional-scale processing. Our novel multi-class variational causal Bayesian inference framework with quadratic variational bounds provides rigorous approximations while ensuring efficiency. By integrating InSAR observations with USGS ground failure models and building fragility functions, our approach separates building damage signals while maintaining computational efficiency through strategic pruning. Evaluation across five major earthquakes (Haiti 2021, Puerto Rico 2020, Zagreb 2020, Italy 2016, Ridgecrest 2019) shows improved damage classification accuracy (AUC: 0.94-0.96), achieving up to 35.7% improvement over existing methods. Our approach maintains high accuracy (AUC > 0.93) across all damage categories while reducing computational overhead by over 40% without requiring extensive ground truth data.

Indonesia is rich in languages and scripts. However, most NLP progress has been made using romanized text. In this paper, we present NusaAksara, a novel public benchmark for Indonesian languages that includes their original scripts. Our benchmark covers both text and image modalities and encompasses diverse tasks such as image segmentation, OCR, transliteration, translation, and language identification. Our data is constructed by human experts through rigorous steps. NusaAksara covers 8 scripts across 7 languages, including low-resource languages not commonly seen in NLP benchmarks. Although unsupported by Unicode, the Lampung script is included in this dataset. We benchmark our data across several models, from LLMs and VLMs such as GPT-4o, Llama 3.2, and Aya 23 to task-specific systems such as PP-OCR and LangID, and show that most NLP technologies cannot handle Indonesia's local scripts, with many achieving near-zero performance.

Southeast Asia (SEA) is a region rich in linguistic diversity and cultural variety, with over 1,300 indigenous languages and a population of 671 million people. However, prevailing AI models suffer from a significant lack of representation of texts, images, and audio datasets from SEA, compromising the quality of AI models for SEA languages. Evaluating models for SEA languages is challenging due to the scarcity of high-quality datasets, compounded by the dominance of English training data, raising concerns about potential cultural misrepresentation. To address these challenges, we introduce SEACrowd, a collaborative initiative that consolidates a comprehensive resource hub that fills the resource gap by providing standardized corpora in nearly 1,000 SEA languages across three modalities. Through our SEACrowd benchmarks, we assess the quality of AI models on 36 indigenous languages across 13 tasks, offering valuable insights into the current AI landscape in SEA. Furthermore, we propose strategies to facilitate greater AI advancements, maximizing potential utility and resource equity for the future of AI in SEA.

Machine learning has the potential to revolutionize passive acoustic monitoring (PAM) for ecological assessments. However, high annotation and compute costs limit the field's efficacy. Generalizable pretrained networks can overcome these costs, but high-quality pretraining requires vast annotated libraries, limiting its current applicability primarily to bird taxa. Here, we identify the optimum pretraining strategy for a data-deficient domain using coral reef bioacoustics. We assemble ReefSet, a large annotated library of reef sounds, though modest compared to bird libraries at 2% of the sample count. Through testing few-shot transfer learning performance, we observe that pretraining on bird audio provides notably superior generalizability compared to pretraining on ReefSet or unrelated audio alone. However, our key findings show that cross-domain mixing which leverages bird, reef and unrelated audio during pretraining maximizes reef generalizability. SurfPerch, our pretrained network, provides a strong foundation for automated analysis of marine PAM data with minimal annotation and compute costs.

In Indonesia, local languages play an integral role in the culture. However, the available Indonesian language resources still fall into the category of limited data in the Natural Language Processing (NLP) field. This is become problematic when build NLP model for these languages. To address this gap, we introduce Bhinneka Korpus, a multilingual parallel corpus featuring five Indonesian local languages. Our goal is to enhance access and utilization of these resources, extending their reach within the country. We explained in a detail the dataset collection process and associated challenges. Additionally, we experimented with translation task using the IBM Model 1 due to data constraints. The result showed that the performance of each language already shows good indications for further development. Challenges such as lexical variation, smoothing effects, and cross-linguistic variability are discussed. We intend to evaluate the corpus using advanced NLP techniques for low-resource languages, paving the way for multilingual translation models.

Shape-morphing capabilities are crucial for enabling multifunctionality in both biological and artificial systems. Various strategies for shape morphing have been proposed for applications in metamaterials and robotics. However, few of these approaches have achieved the ability to seamlessly transform into a multitude of volumetric shapes post-fabrication using a relatively simple actuation and control mechanism. Taking inspiration from thick origami and hierarchies in nature, we present a new hierarchical construction method based on polyhedrons to create an extensive library of compact origami metastructures. We show that a single hierarchical origami structure can autonomously adapt to over 103 versatile architectural configurations, achieved with the utilization of fewer than 3 actuation degrees of freedom and employing simple transition kinematics. We uncover the fundamental principles governing theses shape transformation through theoretical models. Furthermore, we also demonstrate the wide-ranging potential applications of these transformable hierarchical structures. These include their uses as untethered and autonomous robotic transformers capable of various gait-shifting and multidirectional locomotion, as well as rapidly self-deployable and self-reconfigurable architecture, exemplifying its scalability up to the meter scale. Lastly, we introduce the concept of multitask reconfigurable and deployable space robots and habitats, showcasing the adaptability and versatility of these metastructures.

June 6 (Reuters) - When a team of scientists listened to an audio clip recorded underwater off islands in central Indonesia, they heard what sounded like a campfire. Instead, it was a coral reef, teeming with life, according to a study scientists from British and Indonesian universities published last month, in which they used hundreds of such audio clips to train a computer programme to monitor the health of a coral reef by listening to it. A healthy reef has a complex "crackling, campfire-like" sound because of all the creatures living on and in it, while a degraded reef sounds more desolate, life sciences specialist and the team's lead researcher Ben Williams said. The artificial intelligence (AI) system parses data points such as the frequency and loudness of the sound from the audio clips, and can determine with at least 92% accuracy whether the reef is healthy or degraded, according to the team's study published in Ecological Indicators journal. The scientists hope this new AI system will help conservation groups around the world to track reef health more efficiently.

Science  has named nine scientific advances as runners-up for the 2020 Breakthrough of the Year. For 5 decades, scientists have struggled to solve one of biology's biggest challenges: predicting the precise 3D shape a string of amino acids will fold into as it becomes a working protein. This year, they achieved that goal, developing an artificial intelligence (AI) program that predicts most protein structures as accurately as laboratory experiments can map them. Because a protein's precise shape determines its biochemical functions, the new program could help researchers uncover mechanisms of disease, develop new drugs, and even create drought-tolerant plants and cheaper biofuels. Researchers traditionally decipher structures using laborious techniques such as x-ray crystallography and cryo–electron microscopy. But detailed molecular maps only exist for about 170,000 of the 200 million known proteins. Computational biologists have dreamed of simply predicting a protein's structure by modeling the amino acid interactions that govern its 3D shape. But because amino acids can interact in so many ways, the number of possible structures for single protein is astronomical. In 1994, structural biologists launched a biennial competition called the Critical Assessment of Protein Structure Prediction (CASP). Entrants are given amino acid sequences for about 100 proteins with as-yet-unknown structures. Some groups try to predict their structures, while others map the same structures in the lab; afterward, their results are compared. Even in CASP's early years, the predictions for small, simple proteins were on par with experimental observations. But predictions for larger, more challenging proteins lagged far behind. Not anymore. This year, an AI program created by researchers at U.K.-based DeepMind tallied a median score of 92.4 on a 100-point scale, where anything above 90 is considered as accurate as an experimentally derived structure. On the most challenging proteins, the AlphaFold program averaged 87, 25 points ahead of its closest competitor. And because contest rules require competitors to reveal enough of their methods for others to make use of them, organizers say it's only a matter of months before other groups match AlphaFold's success. — Robert F. Service Since the revolutionary genome-snipping tool known as CRISPR burst on the scene in 2012, it has given researchers new power to engineer crops and animals, stirred ethical debates, and earned a Nobel Prize—not to mention Science 's Breakthrough of the Year in 2015. Now, CRISPR is again making waves, scoring its first success in the clinic by treating two inherited blood diseases. People with beta-thalassemia have low levels of the oxygen-carrying hemoglobin protein, leading to weakness and exhaustion; those with sickle cell disease make a defective form of the protein, resulting in sickle-shaped red blood cells that block blood vessels and often cause severe pain, organ damage, and strokes. To treat three sickle cell patients, researchers harvested immature blood cells, known as blood stem cells, from each. They then used CRISPR to disable an “off” switch that—in adults—stops production of the fetal form of hemoglobin, which can counter the effects of the sickling mutation. After the patients received chemotherapy to wipe out their diseased blood stem cells, the CRISPR-treated cells were infused back into their bodies. The patients, treated up to 17 months ago, are now making plentiful fetal hemoglobin, and have not experienced the painful attacks that used to strike every few months, the companies CRISPR Therapeutics and Vertex Pharmaceuticals reported in December. One patient, a young mother of three, says the treatment changed her life. The companies also gave the treatment to seven patients who normally receive blood transfusions for beta-thalassemia. They haven't needed transfusions since, the companies reported in the same paper and meeting presentation. With more testing, the new treatment could rival the success of gene therapies that treat the two diseases by adding hemoglobin DNA to stem cells. But like gene therapy, the CRISPR approach requires high-tech medical care and could cost $1 million or more per patient—putting it out of reach for much of Africa, where most people with sickle cell live. — Jocelyn Kaiser More than 40 years ago, the world's leading climate scientists gathered in Woods Hole, Massachusetts, to answer a simple question: How hot would Earth get if humans kept emitting greenhouse gases? Their answer, informed by rudimentary climate models, was broad: If atmospheric carbon dioxide (CO2) doubled from preindustrial levels, the planet would eventually warm between 1.5°C and 4.5°C, a climate sensitivity range encompassing the merely troubling and the catastrophic. Now, they've finally ruled out the mildest scenarios—and the most dire. Narrowing those bounds has taken decades of scientific advancement. Understanding how clouds trap or reflect heat has been a particular challenge. Depending on their thickness, location, and composition, clouds can amplify warming—or suppress it. Now, high-resolution cloud models, supported by satellite evidence, have shown that global warming thins low, light-blocking clouds: Hotter air dries them out and subdues the turbulence that drives their formation. Longer and better temperature records have also helped narrow the range. Studies of Earth's ancient climate, which estimate paleotemperatures and CO2 levels using ice and ocean sediment cores, suggest how greenhouse gases may have driven previous episodes of warming. And modern global warming has now gone on long enough that surface temperatures, 1.1°C hotter than in preindustrial times, can be used to more confidently project trends into the future. This year, these advances enabled 25 scientists affiliated with the World Climate Research Programme to narrow climate sensitivity to a range between 2.6°C and 3.9°C. The study rules out some of the worst-case scenarios—but it all but guarantees warming that will flood coastal cities, escalate extreme heat waves, and displace millions of people. If we're lucky, such clarity might galvanize action. Atmospheric CO2 is already at 420 parts per million—halfway to the doubling point of 560 ppm. Barring more aggressive action on climate change, humanity could reach that threshold by 2060—and lock in the foreseen warming. — Paul Voosen Everyone loves a good mystery. Take fast radio bursts (FRBs)—short, powerful flashes of radio waves from distant galaxies. For 13 years, they tantalized astronomers keen to understand their origins. One running joke said there were more theories explaining what causes FRBs than there were FRBs. (Currently, astronomers know of more than 100.) Now, cosmic sleuths have fingered a likely culprit: magnetars, neutron stars that fizzle and pop with powerful magnetic fields. Because FRBs are so fast, they must come from a small but intense energy source like a magnetar, which are formed when burned-out stars collapse to the size of a city. But although a handful of FRBs had been traced to particular galaxies, no telescope had sharp enough vision to connect them to an individual magnetar at such great distances. Then, in April, an FRB went off in the Milky Way—close enough that astronomers could examine the scene. The Canadian Hydrogen Intensity Mapping Experiment, a pioneering survey telescope in British Columbia responsible for the discovery of many FRBs, narrowed the source to a small area of sky, which was soon confirmed by the U.S. radio array STARE2. Orbiting observatories sensitive to higher frequencies quickly found that a known magnetar in that part of the sky, called SGR 1935+2154, was acting up at the same time, spewing out bursts of x-rays and gamma rays. Although astronomers studying FRBs believe they have finally found their perpetrator, they still don't know exactly how magnetars produce the radio bursts. They could come from close to the magnetar's surface, as magnetic field lines break and reconnect—similar to the Sun's flaring behavior. Or they could come from farther out, as shock waves slam into clouds of charged particles and generate laserlike radio pulses. Stay tuned for a sequel: Crack theorists are on the case. — Daniel Clery More than 40,000 years ago on the Indonesian island of Sulawesi, a prehistoric Pablo Picasso ventured into the depths of a cave and sketched a series of fantastic animal-headed hunters cornering wild hogs and buffaloes. The age of the paintings, pinned down just 1 year ago, makes them the earliest known figurative art made by modern humans. In 2017, when an Indonesian researcher chanced across the scene, the figures alone told him he had found something special. The animals appear to be Sulawesi warty pigs and dwarf buffaloes, both of which still live on the island. But it was the animallike features of the eight hunters, armed with spears or ropes, that captivated archaeologists. Several of the hunters seem to have long muzzles or snouts. One sports a tail. Another's mouth resembles a bird beak. It's possible the artist was depicting the hunters wearing masks or camouflage, the researchers say, but they may also represent mythical animal-human hybrids. Such hybrids appear in other ancient works of art, including a 35,000-year-old ivory figurine of a lion-man found in the German Alps. Parts of the paintings were covered in white, bumpy mineral deposits known as cave popcorn. Uranium in this popcorn decays at a fixed rate, which allowed researchers to date minerals on top of the pigment to about 44,000 years ago. The cave scene must be at least that old—about 4000 years older than any other known figurative rock art, they reported in late December 2019. It decisively unseats Europe as the first place where modern humans are known to have created figurative art. If the figures do depict mythical human-animal hunters, their creators may have already passed an important cognitive milestone: the ability to imagine beings that do not exist. That, the researchers say, forms the roots of most modern—and ancient—religions. — Michael Price Within days of a racially charged confrontation between a white dog owner and a Black birdwatcher in New York City's Central Park in late May, scientists flocked to Twitter to celebrate—and support—Black nature enthusiasts. The #BlackBirdersWeek hashtag was soon followed by others, in disciplines from neuroscience to physics, all aiming to create community among Black scientists on Twitter, Zoom, and other platforms. “We're few and far between, so having us come together as a conglomerate in one virtual space—it really helped,” says Ti'Air Riggins, a biomedical engineering Ph.D. student at Michigan State University who helped organize #BlackInNeuro week. The social media events took place against the backdrop of the anguished response to police killings in the United States, the Black Lives Matter movement, and discussions within science about the need to create a more equitable, welcoming environment for people of color. Through those discussions, many scientists hoped to reach colleagues who had paid little attention to these issues in the past. “People of color across the board are struggling,” says Tanisha Williams, a botanist at Bucknell University who spearheaded #BlackBotanistsWeek. “It's a systemic problem.” Although it's too early to tell whether the events of this year will spur lasting change, many are hopeful. “This year feels different,” says Shirley Malcom, a senior adviser at AAAS (publisher of Science ) who has worked on diversity, equity, and inclusion issues since the 1970s. “All of a sudden, after George Floyd and everything else that came out after that time, you could at least get people's attention,” she says—adding that many scientists now seem more open to the idea that systemic racism is a problem in their community. “I definitely feel like our voices are being heard, and in a different way [than before],” Williams says. “But it's not going to be a quick fix … we have a long road.” — Katie Langin HIV, like all retroviruses, has a nasty feature that allows it to dodge attack: It integrates its genetic material into human chromosomes, creating “reservoirs” where it can hide, undetected by the immune system and invulnerable to antiretroviral drugs. But where it hides may make all the difference. This year, a study of 64 HIV-infected people who have been healthy for years without antiretroviral drugs reveals a link between their unusual success and where the virus has hunkered down in their genomes. Although the new understanding of these “elite controllers” won't lead directly to a cure, it opens up a novel strategy that may routinely allow other infected people to live for decades without treatment. Many studies have examined elite controllers, who make up about 0.5% of the 38 million people living with HIV. But this new work stood apart in size and scope, comparing integrated HIV in the 64 elite controllers with that in 41 HIV-infected people on treatment. HIV does best when it slots itself within genes. When the cell transcribes the genes, the integrated HIV, or “provirus,” can produce new viruses that infect other cells. If it parks in “gene deserts,” portions of chromosomes that rarely transcribe DNA, the provirus sits around like a fully functioning car stuck in a place that doesn't sell gas. The study found that in the elite controllers, 45% of functioning proviruses resided in gene deserts, compared with just 17.8% for the people on treatment. Presumably, immune responses in the elite controllers somehow cleared proviruses from the more dangerous parking spots. Now, the challenge is to figure out interventions that will train the immune systems of the vast majority of people living with HIV to behave similarly. That new insight suggests long-standing, frustrating attempts to cure people by eliminating HIV reservoirs may be too ambitious an approach. Instead, success may depend on shrinking—and then making peace with—these reservoirs, and minding the old real estate dictum of location, location, location. — Jon Cohen Scientists have spent decades searching for materials that conduct electricity without resistance at room temperature. This year they found the first one, a hydrogen- and carbon-containing compound squeezed to a pressure approaching that at the center of Earth. The discovery is setting off a hunt for room temperature superconductors that work at typical surface pressures; such materials could transform technologies and save the vast amounts of energy wasted when electricity moves through wires. Superconductivity got its start in 1911 when physicist Heike Kamerlingh Onnes found that a mercury wire chilled to 4.2°C above absolute zero, or 4.2 K, conducted electrons without the usual heat-producing friction. In 1986, researchers found the same was true of a family of copper oxide ceramics. Because these superconductors worked above 77 K—the temperature of liquid nitrogen—they spawned a new generation of MRI machines and particle accelerator magnets. There were hints that copper oxides might superconduct at room temperature, but they were never verified. Confirmation now comes from high-pressure physics, in which scientists smash flecks of materials between the flattened points of two diamonds at pressures millions of times higher than those at Earth's surface. With such a diamond anvil, researchers in Germany in 2019 compressed a mix of lanthanum and hydrogen to 170 gigapascals (GPa), yielding superconductivity at temperatures up to 250 K, just under the freezing point of water. This year, researchers in the United States topped that result with a hydrogen, carbon, and sulfur compound compressed to 267 GPa. It conducted without resistance to 287 K, the temperature of a chilly room. So far, the new superconductors fall apart when the pressure is released. But the same isn't true of all high-pressure materials: Diamonds born in the crushing depths of Earth, for example, survive after rising to the surface. Now, researchers hope to find a similarly long-lasting gem for their own field. — Robert F. Service Their eyes are beady and their brains are no bigger than a walnut. But two studies published this year suggest birds have startling mental powers. One reveals that part of the avian brain resembles the human neocortex, the source of human intelligence. The other shows that carrion crows are even more aware than researchers had thought—and may be capable of some conscious thought. In humans, the neocortex consists of horizontal layers laced with interconnected columns of nerve cells, which allow for complex thinking. Bird brains, in contrast, were thought to be arranged in simple clusters of nerve cells. By using a technique called 3D polarized light imaging, neuroanatomists took a closer look at the forebrain of homing pigeons and owls and found that nerves there connect both horizontally—like the layers in the neocortex—and vertically, echoing the columns seen in human brains. Another team of scientists probed this part of the brains of carrion crows—well-known for their intelligence—for clues that they are aware of what they see and do. The researchers first trained lab-raised crows to turn their heads when they saw certain sequences of lights flashing on a computer monitor. Electrodes in the crows' brains detected nerve activity between the moment the birds saw the signal and when they moved their heads. The activity developed even when the lights were barely detectable, suggesting it was not simply a response to sensory input, and it was present regardless of whether the birds reacted. The scientists think the neural chatter represents a kind of awareness—a mental representation of what the birds saw. Such “sensory consciousness” is a rudimentary form of the self-awareness that humans experience. Its presence in both birds and mammals suggests to the researchers that some form of consciousness may date back 320 million years, to our last common ancestor. — Elizabeth Pennisi

Relation Detection is a task to determine whether two entities are related or not. In this paper, we employ neural network to do relation detection between two named entities for Indonesian Language. We used feature such as word embedding, position embedding, POS-Tag embedding, and character embedding. For the model, we divide the model into two parts: Front-part classifier (Convolutional layer or LSTM layer) and Back-part classifier (Dense layer or SVM). We did grid search method of neural network hyper parameter and SVM. We used 6000 Indonesian sentences for training process and 1,125 for testing. The best result is 0.8083 on F1-Score using Convolutional Layer as front-part and SVM as back-part.

Using Facebook Geoinsights, UNICEF was able to confirm internet connectivity was still functioning in the affected area which opened up a new opportunity by working with WhatsApp in the aftermath of... Using Facebook Geoinsights, UNICEF was able to confirm internet connectivity was still functioning in the affected area which opened up a new opportunity by working with WhatsApp in the aftermath of the tsunami to quickly collect needs and provide information to stay alive. Photo credit to UNICEF/UN0240792/Wilander. Rido Saputra, 10 years old, stands in front of his home which was destroyed by a tsunami in Donggala Regency, Central Sulawesi.