Monday, September 28
1,429 new papers appeared. We read every one. These 20 are worth your time.
Today's edition
Picked for broad interest, a real step forward and honest claims, with no field allowed to take over.
Collective order reorganizes dissipation and powers an active engine
These findings establish energetic fluxes as quantitative observables for understanding collective state transitions and demonstrate how collective order can be harnessed for work extraction in active matter.
AI summary
They found that when the spheres become more ordered, they move more efficiently and dissipate energy differently. Energy loss shifts from internal friction to slipping against the surroundings. Using this energy reorganization, they built an active engine that does mechanical work against an external load. The results show that collective order can be used to extract useful work.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepMultidimensional Political Attitudes and Polarization Across 141 Countries
A single left-right axis does not describe political division in most of the world.
AI summary
Researchers combined data from 1.14 million people in 141 countries to measure political attitudes on four dimensions. A single left-right scale does not capture political division in most countries. Different regions divide over different issues: Europe over social liberalism, East Asia and Arab countries over democratic commitment, Latin America over economic left-right, and Africa over anti-elite sentiment. In 39 of 81 countries, the main division would be missed by any left-right scale. Most political disagreement occurs within countries, not between them.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepDirect Light Storage with Controlled Dark Release in a Bipyridine-Based Covalent Triazine Framework
Here, we reveal the light-induced, reversible electron storage properties of a two-dimensional (2D) 2,2′-bipyridine (bpy)-based crystalline covalent triazine framework (CTF), CTF-Bpy.
AI summary
The authors studied a fully organic material that can store energy from light and release it in the dark. The material, called CTF-Bpy, is a crystalline framework made with bipyridine units. Light triggers a proton-coupled electron transfer that stores electrons in the bipyridine units. Stored electrons can be released on demand in the dark to power batteries or produce hydrogen peroxide. With a platinum helper, the stored electrons can also drive hydrogen production in the dark.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
Peer-reviewed journalClaims a big stepReal-world useCo-design of trajectory and morphology for a vertical jump-climbing robot
In this paper we present a 290 g robot that, to our knowledge, is the first to climb vertically by bounding (with an aerial phase).
AI summary
The authors built a 290 gram robot that climbs vertically by bounding, which includes an aerial phase. They used a co-design workflow to jointly optimize the robot's shape and movement path for fast climbing. The optimized path launches the robot upward and reorients it in mid-air using the rear leg as an inertial tail. The robot climbs continuously at 0.375 m/s, can move on the ground, and can transition to vertical surfaces. The authors suggest this work offers design insights and advances toward robots with animal-like climbing agility.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepDual Topological Channels in a Programmable Mechanical Metamaterial
Here we demonstrate that a single mechanical metamaterial can host two independent topological channels governing static and dynamic response within the same architecture.
AI summary
Mechanical metamaterials usually use only one of two different topological mechanisms for robustness. The authors built a single metamaterial that hosts two independent topological channels at once. They used a rotor-chain lattice with two coupled rotational degrees of freedom per unit cell. The two channels can be programmed independently through geometry, angular asymmetry, and stiffness dimerization. Experiments confirmed simultaneous localization of a boundary floppy mode and a finite-frequency domain-wall state.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepGeneration of 10-TW Attosecond X-Ray Pulses in a Free Electron Laser
Here we report the generation of soft-x-ray pulses with a duration of as and peak power on the order of 10 TW at the European XFEL.
AI summary
The researchers generated soft x-ray pulses lasting about 100 attoseconds with peak power around 10 terawatts. They achieved this at the European XFEL using a soft x-ray undulator at high electron energy and five-stage bunch compression. This combination of extremely short pulse duration and high peak power is unprecedented. The millijoule-level attosecond x-ray pulses open a new regime for attosecond science.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepBreaking the Black Box: Byte-Level Boundary Inference of Real-World Antivirus Systems
In this paper, we present AVHunter, the first framework for inferring byte-level decision-critical regions of real-world AV products under a black-box threat model.
AI summary
Current methods for understanding antivirus software only learn simple yes-or-no malware decisions, missing the specific byte regions that trigger those decisions. The authors created AVHunter, the first system that pinpoints exact byte-level regions that control detection in 11 real-world antivirus products. AVHunter's models reproduce antivirus decisions with 97.43% agreement and locate critical byte regions with 85.07% accuracy on average. This work reveals a new type of security knowledge leak with consequences for malware analysis and antivirus defense design.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepReal-world useDistributed laser energy deposition enables fog-resilient optical communications
Here we demonstrate laser-assisted fog clearing under atmospheric-scale propagation conditions over a 140-m optical path using a high-energy, high-average-power picosecond laser and weak focusing to sustain extended multifilamentation.
AI summary
Fog blocks free-space optical communications because tiny water droplets weaken signals by tens of decibels. The authors used a powerful picosecond laser over 140 meters to create a clear channel through fog. This channel, about 2 cm wide, restored up to 12 dB of signal for a separate optical beam. The laser deposits roughly 2 millijoules per meter on average, which may allow scaling to longer weather-resistant links.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintReal-world useJevAdvBench: A Benchmark and Black-Box Attacks for Reinforcement Learning for Calibrated Decisions Models
Building on this, we introduce JevAdvBench, to our knowledge the first adversarial benchmark for RLCD models, with 812 typed questions over 66 scenarios, and a black-box attack suite of 9,744 single-edit variants that each edit one part of a request, with billed input tokens confirming that the edit reached the model.
AI summary
Models like Jev use reinforcement learning to answer questions with probabilities or scores that software acts on automatically. The authors created JevAdvBench, the first benchmark to test how easily these models can be tricked. They compared attacked decisions against the model's own clean decisions rather than outside labels. Adding one unverified opinion flipped 12.1% of decisions, and pushed 38% of confident answers below the 0.8 confidence threshold for human review. Applications using these models should treat the input state as untrusted.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepReal-world useA tidal disruption event in a quasar at redshift 7.19
We report a long-lived nuclear transient in GNz7q, a red quasar at powered by a black hole and hosted by a compact, dusty starburst galaxy.
AI summary
The data fit a model of a star being torn apart by the black hole's gravity, releasing roughly 1.5×10⁵³ ergs of energy. The observations show how black holes episodically feed 700 million years after the Big Bang.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepUncertainty-Aware Federated Learning for Infant Movement Analysis
To address these challenges, we present, to the best of our knowledge, the first federated learning framework for automated infant movement analysis and General Movement Assessment using skeletal motion data.
AI summary
The authors created a new federated learning system to analyze infant movements for early detection of neurodevelopmental disorders. Their uncertainty-aware approach adjusts how much each hospital contributes based on how confident the model is. Results show their federated approach performs much better than local models and nearly matches centralized training.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintReal-world useA Long-Legged, Direct-Drive Monopedal Robot Achieves Exceptional Jump Height
We demonstrate a jumping robot that reaches high (7.6 m) and fast (190 ms stance time) jumps from a single long leg driven by a direct-drive transmission, without elastic energy storage.
AI summary
The robot jumps 7.6 meters high with a single long leg powered by a direct-drive motor, storing no elastic energy. Weighing 281 grams, it achieves the highest jump recorded for an electrically powered system without springs. A new fabric-wrap transmission gives variable mechanical advantage, keeping a small motor at peak power while bracing the lightweight leg. Small propellers at the top control the leg's orientation on the ground and in the air. Tests outdoors included vertical jumps, attitude control, and tilted jumps.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepReprogramming energy system for growth and bioproduction from CO₂ and methanol in Escherichia coli
Integration of the OES-driven CAM* pathway with a rationally rewired native electron transport chain supports growth of the engineered E. coli on CO₂ and methanol, with a doubling time of 8.6 ± 0.3 h and a maximum OD₆₀₀ of 37.5 ± 5.8.
AI summary
The authors built a separate energy system in E. coli bacteria. The engineered bacteria grew with a doubling time of 8.6 hours. The platform also made four chemicals: mevalonate, lactate, 2,3-butanediol, and succinate.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
Peer-reviewed journalClaims a big stepReal-world useBeatGraph: Self-Supervised Heartbeat Graphs for Infant ECG Representations from the Home Environment
We propose BeatGraph, which makes the heartbeat its unit of representation, modeling each 30-second window as a graph of beats.
AI summary
Current ECG models split signals into fixed patches that can break heartbeats apart, which is especially problematic for infants. The authors created BeatGraph, which treats each heartbeat as a unit and models 30-second windows as graphs of beats. They pretrained it on unlabeled infant recordings by predicting masked beats, then fine-tuned it for specific tasks. One model improved sleep-wake detection, infant-state classification, activity-source identification, and affect recognition by 0.076 to 0.158 macro-F1 over baselines. They also released the first public infant ECG dataset from homes, classrooms, and labs: 3,408 hours from 143 infants aged 3 to 11 months.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintReal-world useSN Helios: A Multiply Imaged Type II Supernova Opening Time-Delay Cosmography beyond redshift of 3
We report the discovery and spectroscopic confirmation of SN Helios, a strongly lensed, multiply imaged Type II supernova (SN) at behind the galaxy cluster RXC J0018.5+1626.
AI summary
Astronomers found SN Helios, a multiply imaged Type II supernova at redshift 3.34, using JWST's NIRCam camera. Its spectrum shows hydrogen and helium features confirming it is a hydrogen-rich supernova 27.5 days past peak brightness. Gravitational lensing by a galaxy cluster magnifies it 10 to 20 times and predicts a second image arriving in 2 to 6 years. This time delay may allow measurement of the Hubble constant with a source beyond redshift 3. Without follow-up data, this supernova would appear to be a galaxy at redshift 16, so its spectrum helps identify similar false candidates.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepFederated Targeted Maximum Likelihood Estimation
To fill this gap, our paper introduces the first federated TMLE algorithm.
AI summary
Hospitals and banks often hold data that cannot be shared directly. The authors created the first federated version of targeted maximum likelihood estimation, or TMLE, a statistical method that improves initial estimates. They offer two approaches: one aggregates local gradients step by step, and another lets each site finish its own fit before one exchange. Their protocol sends changes rather than full values and keeps communication error small compared to sampling uncertainty. They note that keeping data local does not guarantee privacy, as sensitive attributes might still be recovered.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepA single-nucleus multi-omic atlas of gene regulation across 21 adult human tissues
Here we present a single-nucleus multi-omic atlas comprising 459,856 transcriptomic and chromatin accessibility profiles from 21 adult human tissues and four donors, including paired measurements from 160,688 nuclei.
AI summary
The authors created a map of gene regulation using 459,856 profiles from 21 adult human tissues. They identified over one million candidate regulatory DNA elements, including 161,270 new ones not in previous databases. Regulatory activity was mainly shaped by cell type but fine-tuned by tissue context. They trained computer models to predict how DNA variants affect chromatin accessibility, finding 18,133 high-effect variants. This atlas offers a framework for understanding gene control and genetic traits across the human body.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepLearning a non-linguistic code for inferred rules from reward
Rewarded for the second's success, the first learns a code carrying rules to three-step transformations training never presents, which new learners acquire.
AI summary
One network invented eight symbols from worked examples; the other applied them to new inputs. Reward for success taught the first network to encode rules for transformations never seen in training. The code has two regimes: reward alone makes messages converge, while expressive pressure keeps them distinct. Success on new rules depends on how much the message reveals about the rule, not message variety.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintExcitation and Identifiability in the 2D Navier-Stokes equations
This paper develops, to the best of our knowledge, the first such excitation theory for a nonlinear partial differential equation.
AI summary
This paper creates the first theory of excitation for a nonlinear partial differential equation. They use a data assimilation method to handle unknown initial states and find candidate viscosities by minimizing loss. At each critical point, either W exceeds a threshold and gives an explicit error estimate, or W is below and observations are insensitive to parameter updates.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintClaims a big stepGenetic architecture of immune-mediated thrombotic thrombocytopenic purpura in Han Chinese individuals
In summary, we reconstructed the genetic architecture of TTP in East Asians and identified structural gatekeepers of autoimmunity, providing novel targets for precision medicine and ethnic-specific risk stratification.
AI summary
Researchers studied the genetic causes of TTP, a rare blood clotting disorder, in 1,344 Han Chinese individuals. They found two genetic regions linked to TTP risk: a known region on chromosome 6 and a newly discovered region on chromosome 14. The chromosome 6 finding involves a specific position in the HLA-DRB1 gene that may affect how the immune system recognizes ADAMTS13 protein fragments. The authors say this work provides new targets for precision medicine and ethnic-specific risk assessment.
Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from
PreprintReal-world use