https://hevolve.ai/research Recent AI and brain-computer interface papers from Nature and arXiv, each explained in plain language. Grounded in what the abstract actually says. en Fri, 17 Jul 2026 17:59:56 GMT https://hevolve.ai/research/knowing-the-self-understanding-the-world-a-dual-cognition-benchmark-fo https://hevolve.ai/research/knowing-the-self-understanding-the-world-a-dual-cognition-benchmark-fo Fri, 17 Jul 2026 17:59:56 GMT Artificial Intelligence arXiv (cs.CV) A drone pilot has to hold two questions in mind at once. Where am I, how fast am I moving, which way am I facing? And separately: what is happening down there, what changed since my last pass? Most tests for vision-language models on aerial footage only ask the second kind of question. They check whether the model can describe a scene or recognise an event, which quietly assumes the hard part is the world and not the aircraft. https://hevolve.ai/research/motionforesight-re-purposing-video-models-for-future-3d-scene-flow-pre https://hevolve.ai/research/motionforesight-re-purposing-video-models-for-future-3d-scene-flow-pre Fri, 17 Jul 2026 17:59:38 GMT Artificial Intelligence arXiv (cs.CV) Watch someone reach for a cup and you already know roughly what happens next. It will be lifted, not slid. A drawer will come out toward you. A lid will rotate shut. You are not calculating physics, you are drawing on a lifetime of having seen objects behave. https://hevolve.ai/research/fvattn-adaptive-sparse-attention-with-runtime-load-balancing-for-video https://hevolve.ai/research/fvattn-adaptive-sparse-attention-with-runtime-load-balancing-for-video Fri, 17 Jul 2026 17:59:32 GMT Artificial Intelligence arXiv (cs.CV) Generating high-resolution video with a diffusion transformer means attending over very long sequences that stretch across both space and time, and self-attention is where the cost concentrates. Sparse attention cuts that bill without retraining by having each head look at only the parts that matter. https://hevolve.ai/research/searching-videos-as-trees-self-correcting-agents-for-grounded-long-vid https://hevolve.ai/research/searching-videos-as-trees-self-correcting-agents-for-grounded-long-vid Fri, 17 Jul 2026 17:59:27 GMT Artificial Intelligence arXiv (cs.CV) Answering a question about a long video means finding the few seconds that actually contain the answer. Recent systems do this by repeatedly cropping: look at a stretch, decide the answer is somewhere in the second half, crop to that, repeat. It narrows quickly, which is the appeal. https://hevolve.ai/research/pagedweight-efficient-moe-llm-serving-with-dynamic-quality-aware-weigh https://hevolve.ai/research/pagedweight-efficient-moe-llm-serving-with-dynamic-quality-aware-weigh Fri, 17 Jul 2026 17:58:29 GMT Artificial Intelligence arXiv (cs.LG) Mixture-of-Experts models are efficient because only a few experts run per token, but every expert's weights still have to be somewhere the GPU can reach. In serving workloads that also carry a large KV cache, those two demands compete for the same memory, and the cache grows with every concurrent conversation. https://hevolve.ai/research/a-blueprint-for-equilibrium-based-differentiable-continuous-variable-t https://hevolve.ai/research/a-blueprint-for-equilibrium-based-differentiable-continuous-variable-t Fri, 17 Jul 2026 17:57:49 GMT Artificial Intelligence arXiv (cs.LG) Machine learning's appetite for energy and its impatience for results are both growing, and this work proposes meeting them with physics rather than more silicon of the usual kind. The idea behind thermodynamic computing is to let a physical system's natural stochastic behaviour perform the computation instead of simulating that behaviour digitally. https://hevolve.ai/research/vision-language-assistant-for-emotional-reactions-to-risky-driving https://hevolve.ai/research/vision-language-assistant-for-emotional-reactions-to-risky-driving Fri, 17 Jul 2026 17:57:27 GMT Artificial Intelligence arXiv (cs.CV) Vision-language models have become good at describing what is happening on a road and reasoning about it. What they almost never consider is how the moment feels to the person in the driver's seat, which is odd, because that is the part determining whether anyone acts on the warning. https://hevolve.ai/research/cluster-aware-matching-via-laplacian-optimal-transport https://hevolve.ai/research/cluster-aware-matching-via-laplacian-optimal-transport Fri, 17 Jul 2026 17:56:51 GMT Artificial Intelligence arXiv (cs.LG) Matching two point clouds usually means finding which point corresponds to which. But in many real settings the points are samples from distributions with genuine cluster structure, and within a coherent region individual points are effectively interchangeable. https://hevolve.ai/research/physics-enhanced-reinforcement-learning-for-real-time-optimal-control https://hevolve.ai/research/physics-enhanced-reinforcement-learning-for-real-time-optimal-control Fri, 17 Jul 2026 17:56:05 GMT Artificial Intelligence arXiv (cs.LG) Reinforcement learning has become an appealing way to control systems too nonlinear or messy for classical methods. Its cost is appetite: the algorithms are sample inefficient, needing an enormous number of interactions before they arrive at a decent control strategy. https://hevolve.ai/research/evaluating-open-weight-llms-for-generating-structured-threat-informati https://hevolve.ai/research/evaluating-open-weight-llms-for-generating-structured-threat-informati Fri, 17 Jul 2026 17:55:19 GMT Artificial Intelligence arXiv (cs.AI) A modern car is a network of computers: sensors, electronic control units, the infotainment system, the telematics unit. A flaw in any of them can put the vehicle, its owner, or the data it holds at risk, and those flaws get catalogued publicly in the CVE database. https://hevolve.ai/research/when-does-muon-help-agentic-reinforcement-learning https://hevolve.ai/research/when-does-muon-help-agentic-reinforcement-learning Fri, 17 Jul 2026 17:49:05 GMT Artificial Intelligence arXiv (cs.AI) Muon holds its own against AdamW when pre-training at scale. Whether that advantage carries into reinforcement learning post-training is a separate question, and this study tests it in the specific setting of sparse-reward agentic RL on ALFWorld with a small Qwen model. https://hevolve.ai/research/behaviour-conditioned-neural-processes-for-adaptive-residential-short https://hevolve.ai/research/behaviour-conditioned-neural-processes-for-adaptive-residential-short Fri, 17 Jul 2026 17:48:22 GMT Artificial Intelligence arXiv (cs.LG) Forecasting electricity demand for a single household is much harder than forecasting it for a city. A city averages out into something smooth. One home is somebody's routine, and routines differ wildly: a night shift worker, a family with school runs, a house that empties every weekend. https://hevolve.ai/research/an-exam-for-active-observers https://hevolve.ai/research/an-exam-for-active-observers Fri, 17 Jul 2026 17:46:23 GMT Artificial Intelligence arXiv (cs.AI) Human seeing is not a snapshot. Your eyes move constantly, and where they go next depends on what you have provisionally concluded from what you have already seen. Psychophysics has argued for decades that this closed loop is not a detail of vision but essential to it. https://hevolve.ai/research/prisa-proactive-infrastructure-lidar-framework-for-intersection-safety https://hevolve.ai/research/prisa-proactive-infrastructure-lidar-framework-for-intersection-safety Fri, 17 Jul 2026 17:35:23 GMT Artificial Intelligence arXiv (cs.LG) Intersections concentrate risk. Vehicles, pedestrians and cyclists all converge with different speeds and sightlines, and the interactions are complex enough that the useful question is not how many crashes happened but which conflicts nearly became one. https://hevolve.ai/research/clife-camera-lidar-fusion-framework-for-edge-deployable-roadside-vru-p https://hevolve.ai/research/clife-camera-lidar-fusion-framework-for-edge-deployable-roadside-vru-p Fri, 17 Jul 2026 17:35:00 GMT Artificial Intelligence arXiv (cs.CV) Combining a camera with LiDAR gives roadside perception the best of both: appearance from one, precise geometry from the other. The difficulty is that pedestrians and cyclists get occluded, lighting changes constantly, and weather does what it likes, so the fusion has to hold up in conditions nobody chose. https://hevolve.ai/research/vtloc-learning-based-tactile-contact-localization-in-visual-point-clou https://hevolve.ai/research/vtloc-learning-based-tactile-contact-localization-in-visual-point-clou Fri, 17 Jul 2026 17:27:08 GMT Artificial Intelligence arXiv (cs.CV) Vision tells a robot what an object is and roughly where. Touch tells it precisely what is happening at the single point where contact occurs. Combining them to answer "where on this object am I touching" is harder than it sounds, because the two views have to be aligned in space to a tolerance neither provides on its own. https://hevolve.ai/research/learning-standard-model-structure-from-lhc-data-with-riemannian-flow-m https://hevolve.ai/research/learning-standard-model-structure-from-lhc-data-with-riemannian-flow-m Fri, 17 Jul 2026 17:23:22 GMT Artificial Intelligence arXiv (cs.LG) Particle collisions at the LHC produce events across an enormous range of energies, and no single Monte Carlo simulation covers the whole span. The work here trains one transformer-based generative model spanning five decades of invariant mass, from below a GeV to the TeV continuum. https://hevolve.ai/research/improving-improved-kernel-pls https://hevolve.ai/research/improving-improved-kernel-pls Fri, 17 Jul 2026 17:19:04 GMT Artificial Intelligence arXiv (cs.LG) Partial least squares is a workhorse for calibration, particularly where you have many correlated measurements and few samples, and the Improved Kernel PLS algorithms are already among the fastest implementations available. This paper speeds up two steps they share. https://hevolve.ai/research/when-do-multi-agent-systems-help-an-information-bottleneck-perspective https://hevolve.ai/research/when-do-multi-agent-systems-help-an-information-bottleneck-perspective Fri, 17 Jul 2026 17:13:44 GMT Artificial Intelligence arXiv (cs.AI) Splitting a hard task across several language model agents has become a popular reflex, and the results are oddly inconsistent. Sometimes a crowd of specialists beats one model working alone, sometimes it does worse, and the field has not had a clean account of when to expect which. https://hevolve.ai/research/toolsciver-multimodal-scientific-claim-verification-with-visual-tool-a https://hevolve.ai/research/toolsciver-multimodal-scientific-claim-verification-with-visual-tool-a Fri, 17 Jul 2026 17:11:50 GMT Artificial Intelligence arXiv (cs.AI) Checking a scientific claim against the paper it came from usually means reading a figure. The number is in a chart, the comparison is in a table, and the qualification that matters is in a caption. Models attempting this fail in three distinguishable ways: they cannot find the decisive visual evidence, they misread structured visuals when they do find them, and they fail to combine what they saw with what they read. https://hevolve.ai/research/a-methodology-for-auditable-trustworthiness-levels-in-ai-lifecycle-gov https://hevolve.ai/research/a-methodology-for-auditable-trustworthiness-levels-in-ai-lifecycle-gov Fri, 17 Jul 2026 17:11:22 GMT Artificial Intelligence arXiv (cs.AI) Governing an AI system is not a single verdict at launch. It means repeatedly asking whether the thing is still trustworthy enough, whether whatever has drifted since last time is tolerable, and being able to show your reasoning to somebody who wants to argue with it. https://hevolve.ai/research/toward-semantic-communication-for-real-time-mobile-3d-reconstruction https://hevolve.ai/research/toward-semantic-communication-for-real-time-mobile-3d-reconstruction Fri, 17 Jul 2026 17:07:40 GMT Artificial Intelligence arXiv (cs.CV) A robot or vehicle building a 3D model of its surroundings as it moves has to send a continuous image stream to a server that does the heavy computation. Unlike offline reconstruction, where you can gather everything first and think later, camera poses and geometry are estimated while the platform is still moving, so multi-view consistency becomes a real-time requirement rather than a post-processing concern. https://hevolve.ai/research/craft-clustering-rubrics-to-diagnose-weak-llm-capabilities-and-generat https://hevolve.ai/research/craft-clustering-rubrics-to-diagnose-weak-llm-capabilities-and-generat Fri, 17 Jul 2026 17:00:38 GMT Artificial Intelligence arXiv (cs.AI) An evaluation that only produces a score has told you where you stand and nothing about what to do next. Most pipelines get as far as identifying weak examples, topics or categories, which sounds diagnostic but is not: it tells you where the model failed while leaving why implicit. https://hevolve.ai/research/rate-utility-frontiers-for-language-encodings-comparing-tokens-bytes-a https://hevolve.ai/research/rate-utility-frontiers-for-language-encodings-comparing-tokens-bytes-a Fri, 17 Jul 2026 16:55:27 GMT Artificial Intelligence arXiv (cs.CL) A language model has to turn text into something it can process, and there are three broad answers: subword tokens, raw bytes, or images of rendered text. Comparisons between them are usually unfair in a way that is hard to see, because each encoding exposes a different amount of actual linguistic content depending on the language. https://hevolve.ai/research/harmonizing-ai-safety-thresholds https://hevolve.ai/research/harmonizing-ai-safety-thresholds Fri, 17 Jul 2026 16:45:30 GMT Artificial Intelligence arXiv (cs.AI) Frontier AI companies publish capability thresholds, the levels at which they say a model becomes dangerous enough to require specific safeguards. The thresholds differ substantially between companies, which creates two problems. https://hevolve.ai/research/the-honest-quorum-problem-epistemic-byzantine-fault-tolerance-for-agen https://hevolve.ai/research/the-honest-quorum-problem-epistemic-byzantine-fault-tolerance-for-agen Fri, 17 Jul 2026 16:43:00 GMT Artificial Intelligence arXiv (cs.LG) Classic consensus protocols are built to survive participants who lie, go silent, or actively collude. The guarantee has a hidden assumption though: everyone outside that faulty set is expected to execute the rules correctly. For ordinary software that is a fair assumption, because code either follows the transition rules or it does not. https://hevolve.ai/research/audio-visual-flamingo-open-audio-visual-intelligence-for-long-and-comp https://hevolve.ai/research/audio-visual-flamingo-open-audio-visual-intelligence-for-long-and-comp Fri, 17 Jul 2026 16:41:15 GMT Artificial Intelligence arXiv (cs.CV) Most audio-visual language models are built and tested on short clips, which quietly avoids the thing that makes real video hard. In a long recording the sound and the picture drift in and out of agreement, evidence for an answer can be minutes apart, and what you heard early on changes how you should read what you see later. https://hevolve.ai/research/attention-guided-saliency-maps-for-interpreting-visualization-literacy https://hevolve.ai/research/attention-guided-saliency-maps-for-interpreting-visualization-literacy Fri, 17 Jul 2026 16:40:23 GMT Artificial Intelligence arXiv (cs.CV) Vision-language models are increasingly asked to read charts and draw conclusions, which makes it worth knowing how they arrive at an answer. A model that reads the correct bar and one that guesses from the caption produce identical output and deserve very different amounts of trust. https://hevolve.ai/research/understanding-reasoning-from-pretraining-to-post-training https://hevolve.ai/research/understanding-reasoning-from-pretraining-to-post-training Fri, 17 Jul 2026 16:31:58 GMT Artificial Intelligence arXiv (cs.AI) Reinforcement learning has become the standard way to sharpen a language model's reasoning after the main training run. Yet the two stages are usually studied apart, which leaves two basic questions open: how do pretraining decisions like model size and data change what RL can add, and what is RL really doing to the model. https://hevolve.ai/research/how-do-vlms-fail-vision-operation-misalignment-in-compositional-vqa https://hevolve.ai/research/how-do-vlms-fail-vision-operation-misalignment-in-compositional-vqa Fri, 17 Jul 2026 16:25:03 GMT Artificial Intelligence arXiv (cs.CV) Asking a model "is the red mug to the left of the laptop" is not one question but several: find the mug, confirm it is red, locate the laptop, work out the spatial relation. Vision-language models post respectable aggregate scores on this and nobody has looked closely at how the failures are distributed. https://hevolve.ai/research/dadiff-diffusion-driven-cross-domain-policy-adaptation-for-reinforceme https://hevolve.ai/research/dadiff-diffusion-driven-cross-domain-policy-adaptation-for-reinforceme Fri, 17 Jul 2026 16:20:08 GMT Artificial Intelligence arXiv (cs.AI) A policy learned in simulation rarely survives contact with the real thing, because the two worlds do not behave identically. Friction differs, delays differ, the mass is not quite what the model assumed. This is the dynamics mismatch, and it is why so much reinforcement learning stays in the simulator. https://hevolve.ai/research/neural-spectroscopy-of-alphafold2-reveals-encoded-protein-conformation https://hevolve.ai/research/neural-spectroscopy-of-alphafold2-reveals-encoded-protein-conformation Fri, 17 Jul 2026 16:18:07 GMT Artificial Intelligence arXiv (cs.LG) AlphaFold2 is normally treated as a machine: sequence goes in, structure comes out, and the 93 million parameters in between are plumbing. The proposal here is to treat those parameters as a scientific object in their own right. https://hevolve.ai/research/controlling-implicit-shortcut-reliance-in-l2-spoken-english-auto-marke https://hevolve.ai/research/controlling-implicit-shortcut-reliance-in-l2-spoken-english-auto-marke Fri, 17 Jul 2026 16:15:49 GMT Artificial Intelligence arXiv (cs.CL) Automatic marking of spoken English increasingly feeds audio or text straight into a large model rather than extracting hand-designed features first. That removes a bottleneck and removes a safeguard at the same time, because those features encoded what assessors believed should count. https://hevolve.ai/research/pick-to-learn-calibration-of-an-mpc-policy-for-an-origin-to-destinatio https://hevolve.ai/research/pick-to-learn-calibration-of-an-mpc-policy-for-an-origin-to-destinatio Fri, 17 Jul 2026 16:14:29 GMT Artificial Intelligence arXiv (cs.LG) The example is an aircraft flying from one point to another through uncertain crosswinds while avoiding a zone of poor connectivity, controlled by a model predictive policy. The paper is explicit that the aircraft is a vehicle for the method rather than the subject. https://hevolve.ai/research/physics-based-deep-spatiotemporal-hyperlocal-radar-nowcasting-with-a-m https://hevolve.ai/research/physics-based-deep-spatiotemporal-hyperlocal-radar-nowcasting-with-a-m Fri, 17 Jul 2026 16:09:48 GMT Artificial Intelligence arXiv (cs.LG) Predicting rain in the next ten to ninety minutes is its own problem, separate from forecasting tomorrow's weather. Conventional high-resolution numerical models can do it, but they need repeated data assimilation, initialisation and spin-up before they produce anything, and that lead time is exactly what you do not have when you are deciding whether a road is about to flood. https://hevolve.ai/research/adaptive-contrast-enhancement-and-optimised-feature-matching-for-roots https://hevolve.ai/research/adaptive-contrast-enhancement-and-optimised-feature-matching-for-roots Fri, 17 Jul 2026 16:04:15 GMT Artificial Intelligence arXiv (cs.CV) Palm-vein recognition is among the more secure biometrics precisely because the pattern is under the skin. It cannot be lifted from a glass or captured from a photograph the way a fingerprint or a face can, and it needs a living hand present. https://hevolve.ai/research/hcig-a-hierarchical-cross-modal-incongruity-graph-network-for-multimod https://hevolve.ai/research/hcig-a-hierarchical-cross-modal-incongruity-graph-network-for-multimod Fri, 17 Jul 2026 16:02:46 GMT Artificial Intelligence arXiv (cs.AI) Sarcasm rarely lives in the words alone. A caption reading "what a beautiful day" over a photo of torrential rain means the opposite of what it says, and neither the text nor the image carries that meaning by itself. It exists in the mismatch. The same is true of much online bullying, where an innocuous phrase turns cruel next to a particular picture. https://hevolve.ai/research/joynexus-service-oriented-multi-tenant-post-training-for-vla-models https://hevolve.ai/research/joynexus-service-oriented-multi-tenant-post-training-for-vla-models Fri, 17 Jul 2026 15:58:20 GMT Artificial Intelligence arXiv (cs.AI) Models that turn what a robot sees into what a robot does have to be retrained constantly, because every simulator, every robot body and every task objective is a little different. That retraining has to run somewhere, and the way compute is currently sold makes it awkward. https://hevolve.ai/research/frontier-language-models-struggle-to-copy-text-can-be-better-viewed-in https://hevolve.ai/research/frontier-language-models-struggle-to-copy-text-can-be-better-viewed-in Fri, 17 Jul 2026 15:56:52 GMT Artificial Intelligence arXiv (cs.CL) Frontier models solve hard reasoning problems in seconds and, the authors show, fail at something far simpler: reproducing an input string exactly, when that string sits comfortably inside the context window. The gap between those two facts is the finding. https://hevolve.ai/research/llm-powered-agentic-ai-for-5g6g-networks-a-tutorial-and-survey-on-arch https://hevolve.ai/research/llm-powered-agentic-ai-for-5g6g-networks-a-tutorial-and-survey-on-arch Fri, 17 Jul 2026 15:47:23 GMT Artificial Intelligence arXiv (cs.AI) Running a mobile network has been rule-based automation: conditions are detected, predefined responses fire. Agentic AI proposes something different, systems given goals that decide for themselves how to reach them, and next-generation networks are an obvious candidate because their configuration space long ago outgrew what people can hand-tune. https://hevolve.ai/research/spatial-normalization-for-cross-domain-retinal-layer-segmentation-in-o https://hevolve.ai/research/spatial-normalization-for-cross-domain-retinal-layer-segmentation-in-o Fri, 17 Jul 2026 15:47:18 GMT Artificial Intelligence arXiv (cs.AI) Optical coherence tomography produces cross-sectional images of the retina, and separating out its layers is the first step before you can measure anything. Those measurements are drawing interest well beyond eye care, including as possible markers in neurodegenerative disease. https://hevolve.ai/research/when-model-merging-rivals-joint-multi-task-reinforcement-learning-a-ta https://hevolve.ai/research/when-model-merging-rivals-joint-multi-task-reinforcement-learning-a-ta Fri, 17 Jul 2026 15:41:09 GMT Artificial Intelligence arXiv (cs.AI) Model merging is offered as a substitute for joint multi-task training, and the authors notice something awkward about the evidence. Merging is normally applied to independently released agents precisely because no joint model exists, which means the comparison against the thing it claims to replace is almost never run. https://hevolve.ai/research/artchart-a-benchmark-for-faithful-artistic-chart-generation-with-integ https://hevolve.ai/research/artchart-a-benchmark-for-faithful-artistic-chart-generation-with-integ Fri, 17 Jul 2026 15:38:33 GMT Artificial Intelligence arXiv (cs.CV) A beautiful chart is worth a lot, because people remember it. Generating one automatically means satisfying several constraints at the same time, and they pull against each other: the numerical geometry has to stay accurate, text has to render exactly rather than approximately, each label has to attach to the right mark, and the whole thing has to hold a coherent artistic style. https://hevolve.ai/research/frontier-ai-performance-across-the-business-disciplines-a-case-grounde https://hevolve.ai/research/frontier-ai-performance-across-the-business-disciplines-a-case-grounde Fri, 17 Jul 2026 15:34:51 GMT Artificial Intelligence arXiv (cs.AI) Benchmark scores keep climbing, and what they measure is narrower than the headlines suggest: factual recall, focused question answering, mathematics, coding, tool use. All real capabilities, and none of them is what a white-collar professional actually spends the day doing. https://hevolve.ai/research/multi-modal-semantic-segmentation-of-electrolyzer-components-for-susta https://hevolve.ai/research/multi-modal-semantic-segmentation-of-electrolyzer-components-for-susta Fri, 17 Jul 2026 15:34:20 GMT Artificial Intelligence arXiv (cs.CV) Recycling an electrolyzer means telling its materials apart well enough for a machine to take it to pieces. That is harder than it sounds, because the materials look alike, overlap spectrally, come in irregular shapes, and appear in wildly unequal amounts. https://hevolve.ai/research/deep-and-probabilistic-models-for-gene-regulatory-network-inference https://hevolve.ai/research/deep-and-probabilistic-models-for-gene-regulatory-network-inference Fri, 17 Jul 2026 15:31:05 GMT Artificial Intelligence arXiv (cs.LG) A gene regulatory network describes which transcription factor proteins switch which genes on or off. Reconstructing one from genome-wide measurements is a long-standing problem, and this work is as interested in how the field evaluates its answers as in the answers themselves. https://hevolve.ai/research/loop-the-loopies https://hevolve.ai/research/loop-the-loopies Fri, 17 Jul 2026 15:28:43 GMT Artificial Intelligence arXiv (cs.AI) A looped transformer runs the same layers repeatedly rather than stacking more of them, which is appealing because it reuses parameters instead of multiplying them. The idea has faced a stubborn empirical problem: given N times more pre-training compute, making the model N times bigger has usually beaten looping it N times. https://hevolve.ai/research/deluge-towards-continental-scale-daily-pluvial-flood-damage-prediction https://hevolve.ai/research/deluge-towards-continental-scale-daily-pluvial-flood-damage-prediction Fri, 17 Jul 2026 15:27:47 GMT Artificial Intelligence arXiv (cs.LG) Pluvial flooding is the kind that comes straight from rainfall, when water arrives faster than the ground and drains can take it, rather than from a river bursting or the sea coming in. It accounts for 45% of National Flood Insurance Program claims in the United States, and it is the hardest of the three to predict. https://hevolve.ai/research/sciforge-an-ai-native-multimodal-workbench-for-scientific-discovery https://hevolve.ai/research/sciforge-an-ai-native-multimodal-workbench-for-scientific-discovery Fri, 17 Jul 2026 15:13:03 GMT Artificial Intelligence arXiv (cs.AI) A single piece of research is scattered across papers, code, datasets, odd scientific file formats, model outputs, figures, drafts, and decisions the team made in conversation somewhere. General-purpose AI assistants can help with any one of those, but they rarely hold them together as a coherent state you could later audit. https://hevolve.ai/research/revisiting-data-driven-dynamic-security-assessment-with-a-tabular-foun https://hevolve.ai/research/revisiting-data-driven-dynamic-security-assessment-with-a-tabular-foun Fri, 17 Jul 2026 15:06:11 GMT Artificial Intelligence arXiv (cs.AI) Grid operators need to know, before anything goes wrong, which credible faults would actually destabilise the system. Machine learning does this quickly, but the usual setup has an awkward shape: you train, tune and maintain a separate model for every contingency on a long list, and each one needs a large labelled database behind it.