Science & Mathematics

Training generative models on synthetic data causes loss of diversity and a decrease in performance, a phenomenon known as model collapse. We study this degeneration spectrally in a controlled Markov setting. The eigenvalue spectrum of the learned transition operator offers a direct view of the process, since the subdominant eigenvalues set how fast the chain mixes and therefore how much of the state space it continues to explore. We find that sampling at temperature $T_s < 1$ acts on the operator like repeated sharpening, pushing it toward a deterministic system: the subdominant eigenvalues approach the unit circle and the spectral gap narrows. When $T_s < 1$, even data accumulation does not appear to arrest this drift. We then analyze Dreaming Learning, a regulariser that, in parallel with regular training, adds a loss term on the model's own samples drawn at a raised temperature $T_d > 1$. Because high-temperature samples flatten the predicted distribution, training on them keeps the entries of the learned operator from becoming vanishingly small, which in turn keeps the eigenvalues bounded away from the unit circle. Measurements of the spectra over 5000 generations are consistent with this picture, and the mitigation effect persists in more complex settings such as LSTM language models and a 124M-parameter GPT-2.

Recently, two-dimensional goldene, a single-atomic-layer gold sheet, has been experimentally realized and is expected to be a promising material for next-generation devices, offering new possibilities beyond three-dimensional bulk gold. Because quasi-one-dimensional derivatives can exhibit properties markedly different from their two-dimensional counterparts, we investigate the stability, electronic, and magnetic properties of goldene-based nanoribbons (GOLDNRs) using first-principles computational methods. Our findings demonstrate that the electronic structure of these nanoribbons depends strongly on both edge chirality and ribbon width. Depending on the boundary configuration, the systems can exhibit semiconducting behavior with electronic band gaps of approximately 0.77 eV and 0.04 eV, or retain metallic characteristics. Additionally, the edge regions act as electronically active sites that may significantly enhance chemical reactivity relative to the central regions. As the ribbon width increases, the metallic characteristics of the infinite goldene sheet are progressively recovered, while the influence of quantum confinement and edge effects on the central region diminishes. These results provide important insights into tailoring the physical properties of goldene through lateral confinement, highlighting its potential for applications in nanoscale field-effect transistors, spintronic devices, and catalysis.

Deploying voltage source converter (VSC)-based multi-terminal DC (MTDC) networks in existing AC grids offers a promising solution for alleviating transmission congestion, enhancing power-transfer capability, and improving operational economy. However, constructing an MTDC network requires participation from merchant investors. Therefore, the transmission system operator (TSO) responsible for the AC grid must provide an adequate economic incentive that enables merchant investors not only to recover the MTDC investment cost but also to earn a reasonable return commensurate with their risk and expectations. An insufficient incentive may fail to attract merchant investment, whereas an excessive incentive may unnecessarily reduce the economic benefits retained by the TSO. Determining an appropriate incentive that balances these competing interests therefore becomes a critical issue. To address this issue, this paper formulates a sharing-rate-parameterized mixed-integer second-order cone programming (MISOCP) model that jointly optimizes VSC siting and sizing while coordinating AC/DC system operation. A straightforward search procedure is then employed to identify the benefit-sharing rate that meets the investor's expectations while preserving sufficient benefits for the TSO. Case studies demonstrate the operating-cost benefits of MTDC integration and validate the proposed search strategy for seeking such a benefit-sharing rate.

Language 3D Gaussian Splatting enables open-vocabulary scene understanding by associating Gaussian primitives with language features. These semantic attributes increase model storage, while most existing 3DGS compression methods are designed mainly to preserve RGB rendering quality. One straightforward approach is to apply sparsification and quantization sequentially. However, pruning based only on rendering quality can remove Gaussians that have small RGB contributions but are important for semantic prediction. Once removed, their semantic information cannot be recovered by later quantization. We propose ClunoGS, a compact framework for Language 3DGS that combines semantics-aware Gaussian sparsification and vector quantization in a unified constrained objective. Semantic supervision is used during sparsification, and the objective jointly considers rendering quality, semantic accuracy, and compactness. We optimize the unified objective with alternating primal, sparsification, quantization, and dual updates. We evaluate ClunoGS on open-vocabulary localization and segmentation using the LERF and ScanNet-10 benchmarks. Under aggressive compression, our unified, semantics-aware formulation substantially alleviates the degradation of semantic representations caused by sequential compression and semantics-agnostic Gaussian sparsification. Ablations further show that jointly coordinating sparsification and quantization better preserves semantic accuracy than applying them independently.

LLM agents with similar success rates can follow different action sequences. An Agent Behavior Map represents LLM-harness combinations using centered action log-likelihoods on shared reference trajectories, without requiring new environment rollouts from the target agent. We derive error bounds relating squared coordinate distances to trajectory KL under explicit assumptions, and across eight cross-scored models the distances track the directly measured trajectory KL in rank order. Across 100 LLMs and four environments, ALFWorld-based coordinates support performance prediction and identify behavioral similarities. In ALFWorld, these similarities persist after controlling for success rates and model size, including across model families. As an extension, harness-component displacements approximately compose, with accuracy varying across components and environments.

Antenna diversity in Internet of Things (IoT) devices is explored to address challenges associated with fixed device orientation and interference from nearby metallic objects, which can degrade wireless network performance. Integrating multiple antennas with different polarizations can significantly enhance signal quality and network reliability. Experiments demonstrated significant improvements in link quality, with gains exceeding 30 dB in received signal strength for specific scenarios. Already the addition of a second PCB-based antenna alone, resulted in a 17 dB improvement. These findings suggest that the added complexity in hardware and software to incorporate antenna diversity could well be justified by the resulting enhancements in network stability for real-world applications.

Understanding large, complex, and rapidly evolving code bases remains a core challenge for software engineering at scale, hindering onboarding, maintenance, refactoring, and impact analysis. Providing this information to a Large-Language Model (LLM) is another hurdle, as the LLM must understand not only the required code method but also its associated methods in the grand scheme of the repository. We propose Code Comprehensor, a hybrid retrieval-reasoning framework combining (i) CodeGraph, a code dependency callgraph, (ii) CodeScribe, which uses code-aware LLMs to generate hierarchical documentation across repository call graphs, and (iii) CodeContrast, a CLIP-style dual-encoder model trained on CodeScribe output to learn aligned text-code embeddings. CodeContrast enables fast, semantically precise retrieval of relevant code-context neighborhoods, which serve as grounded evidence for an MCP-based LLM orchestrator answering domain-specific queries. Evaluation on OpenCV and a proprietary 5M-LOC repository shows that CodeScribe-derived supervision significantly improves cross-modal retrieval in text-to-code accuracy over baseline approaches. Richer textual supervision (hierarchical documentation, parent/sibling context, and LLM-generated synthetic questions) consistently enhances performance across both repositories. LLM-as-a-judge evaluation using the popular RAGAS of the Code Comprehensor output yields above 90% faithfulness scores.

This study introduces GuideEvolve, a self-evolving agent framework incorporating expert guidance and three autonomous regulatory mechanisms, collectively termed the 3Ts: Think Fast and Slow, Throwback, and Termination by choice. The framework addresses critical limitations in extant self-evolving agent paradigms through dual innovations: (1) a task-specific expert-evaluation-driven evolutionary pipeline, and (2) the integration of 3Ts mechanisms to facilitate cognitive optimization, performance recovery via temporal rollback, and adaptive termination protocols. Empirical validation across the MGSM and MMLU benchmarks demonstrates that GuideEvolve attains performance metrics comparable to current state-of-the-art models (89.8% accuracy versus Gödel Agent's 90.6% on MGSM; 89.1% versus 87.9% on MMLU), while exhibiting notably reduced variance (25-59%) and enhanced robustness. Furthermore, the framework achieves a 100% evolution success rate, superior to the baseline's 82%, diminishes required optimization iterations by approximately 50% (from 30 iterations in baseline to 15.1 and 13.4 across tasks), and provides improved stability for production deployment.

Adversarial training derives updates from surrogate losses, but source-loss descent does not directly characterize an update's effect on adversarial margins. We introduce ExGrad (extrapolative gradient recomposition), which constructs clean-input, adversarial-input, and adversarial-weight-perturbed (AWP) gradients. Virtual steps measure their margin responses, and a globally solvable three-variable problem selects coefficients over a bounded affine domain without nonnegativity constraints. Across affine radii on CIFAR-10, negative clean coefficients rapidly predominate and increase in magnitude with the radius. Cross-zero interventions show that setting the clean coefficient to zero is insufficient: mild subtraction improves PGD-20, while stronger subtraction trades clean accuracy for further PGD-20 gains. In CIFAR-10 WRN-28-4 controls, none of the three candidate-removal variants matches full ExGrad. Projection diagnostics show that negative clean coefficients usually coexist with positive net clean-axis projections, although both decrease as the radius grows. Norm-matched controls indicate that update magnitude alone does not explain the moderate-radius gains. Experiments on CIFAR-10, CIFAR-100, and ImageNet-100 show radius-dependent improvements over AWP. On CIFAR-100, ExGrad at $R=0.8$ improves all five reported metrics, including clean and AutoAttack accuracy by 3.84 and 1.75 percentage points. On ImageNet-100, the same radius improves all four robust metrics over both AWP and PGD-AT. These results identify clean-gradient subtraction as a response-revealed update preference whose strength and clean-robust operating point are controlled by the affine radius.

Language models have shown strong potential for intrusion detection, but their high memory and computational demands make them difficult to deploy in resource-constrained IoT environments. In this work, we propose a compact Transformer-based architecture that compresses the input into a small latent space before processing it with a minimal Transformer core. We analyze nine compression configurations, compare parameter sharing with layer stacking, and evaluate our approach against SecurityBERT on two IoT intrusion detection datasets, Edge-IIoTset and CICIoT2023. With only 306k non-embedding parameters and 1.6M parameters in total (6MB), our model is about 7x smaller than SecurityBERT (11.2M parameters, 45MB). With the same privacy-preserving encoding, it outperforms SecurityBERT by about 2 points of macro F1-score on Edge-IIoTset and is within 0.7 points on CICIoT2023. It is also about 89x faster on a simulated edge device. We further show that a single Transformer layer is sufficient, and that parameter sharing matches layer stacking with up to 66% fewer parameters. These findings show that lightweight Transformer-based architectures can provide a practical and accurate solution for intrusion detection in resource-constrained IoT settings

The non-preemptive dial-a-ride problem (DaRP) asks for a minimum-cost route of a capacitated vehicle that serves a collection of pickup-and-delivery requests, without allowing objects to be transferred or temporarily unloaded. We study this problem and the closely related minimum-ratio Steiner $k$-Forest (S$k$F) in general symmetric metrics on $n$ locations. We develop a randomized polynomial-time algorithm with expected approximation ratio $\mathcal{O}(n^{1/4}\log^{11}n)$ for the minimum-ratio S$k$F. Our algorithm combines a deterministic approximation on hierarchically well-separated trees with randomized metric embeddings. Using this algorithm, we obtain expected $\mathcal{O}(n^{1/4}\log^{12}n)$-approximations for both the non-preemptive DaRP and the S$k$F. These guarantees improve on the previous $n^{0.3+\varepsilon}$-approximation ratios for the S$k$F and the non-preemptive DaRP, where $\varepsilon>0$ is any fixed constant. Complementing these algorithmic results, we establish the first superconstant inapproximability bounds for the non-preemptive DaRP. Under the Exponential Time Hypothesis (ETH), no deterministic polynomial-time algorithm achieves an $x^{1/(\log\log x)^c}$-approximation for some constant $c>0$, where $x$ denotes either the number of locations $n$ or the vehicle capacity $k$. Under the quantitative Gap-ETH, we rule out every deterministic polynomial-time $x^{o(1)}$-approximation. These bounds hold even on unit-edge stars with $n=(2+o(1))k$.

Empirical exchange-correlation functionals are fitted on fixed orbitals but used self-consistently, so better energies often come with worse densities. We show that the Brillouin residual of a trained functional on the Kohn-Sham orbitals of an accurate density controls, to first order, both the self-consistent density shift and the gap between fitted and self-consistent energies. Penalizing it requires no self-consistent-field solution during training. Trained on 6 energies and 11 CCSD densities, a neural meta-GGA surpasses SCAN in both atomization energies and densities of unseen molecules.

This paper addresses the problem of creating immersive volumetric videos from sparse and low-resolution capture resources. Conventional dense multi-view reconstruction methods can produce high-quality volumetric videos, but their reliance on costly capture setups makes them impractical in real-world scenarios. Recent approaches use generative models to relax capture requirements, but still suffer from limited resolution and 4D inconsistency. To overcome these constraints, we propose RealMe, a unified framework built around a high-resolution video diffusion model that enhances limited captures into high-resolution (2K), 4D-consistent videos for volumetric video reconstruction. Specifically, RealMe constructs auxiliary data from capture videos and encodes them into latents along with layout-aware positional embeddings. The video diffusion model adapts factorized spatio-temporal attention to the multi-view capture layout, denoises these latents across viewpoints and time, and decodes them into enhanced videos to refine the 4D representation. This adaptation helps the model capture temporal dynamics and multi-view relationships, leading to more 4D-consistent generation. To reduce the data gap between generated videos and real captures, we apply a latent frequency loss that provides frequency-domain supervision in latent space. Extensive experiments on the public DNA-Rendering benchmark and our self-captured RealVV dataset demonstrate that RealMe is able to create immersive volumetric videos and outperforms state-of-the-art methods in visual fidelity and 4D consistency.

Enterprise agents act on documents. Before an agent can follow a policy, answer a compliance question or file a ticket, some ingestion layer must turn a Word file or PDF into tokens - and that layer is where an agent deployed in the wild silently loses information. We show that one layer is lost by every mainstream converter - the meaning an author encodes in how text looks - and that the loss makes agents confidently wrong rather than visibly incomplete. We build a task from 76 real policy documents: recover the requirements the author singled out, scored against the file's own OOXML properties by a deterministic matcher with no LLM judge. Formatting-blind ingestion collapses from micro F1 0.690 to 0.150 (docling) and 0.075 (plain text), a gap that replicates across six answering models from two vendors - five Claude models spanning the Opus, Sonnet and Haiku tiers, and OpenAI's GPT-5.6. An agent that cannot see the cue guesses, and docling's false-positive rate on requirements the author did not flag is four times ours. A span-level leave-one-out localises the loss to two cues - highlight (+0.302) and colour (+0.262), both Holm-adjusted p < 1e-4 - while bold, italic, monospace and strikethrough each sit within +/-0.003 of zero. The same asymmetry appears in PDF. Colour is also the cue no mainstream converter preserves and no parsing benchmark scores, and it is common: 32.8% of 6,460 documents carry a deliberate colour or highlight, 10.3% on an obligation. A pre-registered test shows that typing a cue for the agent adds +0.085 beyond merely preserving it (Holm-adjusted p = 0.049). We validate the labels against rendered page images through an independent modality (kappa = 0.72) and bound what wording alone can achieve (AUC 0.664). For agent builders the practical finding is that ingestion, not reasoning, is where this reliability is won or lost, and that the fix is cheap and measurable.

The Pandora spacecraft is a smallsat selected in NASA's inaugural class of Pioneers missions. The aim of the Pandora mission is to disentangle stellar activity signals from exoplanet atmospheric transmission spectroscopy measurements. Pandora accomplishes this goal with its combined visible photometry and NIR spectroscopy that observe each target in the prime mission for 10 separate transit observations, each spanning 24 hours at a time. Targets were carefully selected for the Pandora prime mission. Pandora is specifically studying planets around low-mass stellar hosts (K and M dwarf stars) spanning a range of rotation periods and activity levels. The planets range in size from 1.5 - 15 Earth radii and orbital periods ranging from 1.7 - 11.5 days. In addition to the 10 transit visits planned for each of the 20 prime mission targets, Pandora is observing a number of auxiliary targets encompassing a variety of science topics. This includes topics such as stellar activity and young systems, transit timing variations of exoplanets, brown dwarf compositions, exoplanet phase curves, etc. We will discuss the selection of these targets, how they have been scheduled to fill in gaps in prime mission operations, and early science results. If the mission is extended, there is further opportunity for accomplishing additional science. Here we will discuss the possible options for extended mission operations and science.