Papers

Modern learning problems increasingly leverage data from heterogeneous sources to improve robustness and generalizability under distribution shift, often with high-dimensional parameters admitting low-rank structure. Existing group distributionally robust optimization (group-DRO) approaches optimize against adverse source mixtures, but extending this principle to low-rank matrix estimation with a broad class of losses leads to a challenging nonconvex-concave problem for which methods and theory remain limited. We propose Prior-Informed generalizable Low-rank representation Learning via Adversarial Reweighting (PILLAR), which smooths the group-DRO objective by regularizing adversarial source weights toward a user-specified reference distribution while learning a shared low-rank representation. By tempering extreme adversarial reweighting, PILLAR improves estimation stability and reduces the potential conservativeness of group-DRO. Using Burer-Monteiro factorization, we optimize the nonconvex-concave objective by bifactored gradient descent with mirror ascent. We establish linear convergence to a statistical neighborhood of the population saddle point and derive an error bound that separates low-rank estimation from adversarial reweighting. We further show that PILLAR improves on vanilla group DRO when the latter's excess conservativeness dominates source heterogeneity and prior misspecification. Simulations and an electronic health record study of suicide-attempt risk demonstrate improved worst-source and transported predictive accuracy.

Computational coarse-graining aims to replace a sequence of updates with a cheaper module while preserving task performance. We present a continuation-relative evaluation framework that assesses replacements through the computation that follows them, using downstream task quality and total execution cost as joint criteria. The framework distinguishes boundary representations, summaries of skipped computation, executed operations, and downstream loss. It makes interfaces and permitted deployment information explicit, combines fine-path references with directly trained controls using the same shortened architecture, and accounts for selection overhead. An exact loss decomposition separates replacement effects at a fixed downstream action from effects of changing that action. A finite construction shows how retained information can reduce the stipulated cost of zero-error prediction under fixed information and controller restrictions. Experiments in two frozen pretrained language models show that representation fidelity and task loss can favor different fitting objectives. In a prespecified comparison on a 1.7-billion-parameter model, next-token-loss training lowers in-distribution negative log-likelihood by 0.1901 nats per scored token compared with reconstructing the skipped blocks' output, using the same optimizer and update count. This ranking reverses in all nine paired comparisons under distribution shift. In a recursive expert prototype, the jointly adapted candidate lowers task error relative to frozen-bank and router-only baselines, but trails directly trained controls, and the tested input-only selectors fail the joint quality-cost criteria. Together, the protocol and studies provide explicit criteria and diagnostics for distinguishing useful computational replacement from local approximation gains within specified interfaces and evaluation settings.

Characterizing subsurface fracture networks is challenging because geometry and connectivity are sparsely observed and uncertain away from measurements. We develop a latent diffusion framework for unconditional and observation-conditioned generation of two-dimensional (2D) and three-dimensional (3D) fracture networks. Fractures and the surrounding matrix are represented as binary fields and compressed by a variational autoencoder into compact spatial latent representations, where an Elucidated Diffusion Model learns their distribution. The 3D model operates on native volumetric fields, with a reconstruction objective designed to preserve sparse, thin fracture structures. Sparse borehole observations are incorporated through a conditioning encoder, classifier-free guidance, and diffusion posterior sampling. Generated ensembles reproduce key fracture-network statistics, including fracture area/volume fraction, two-point correlation, and chord-length distributions. Reconstructed fracture counts and gas-production responses are comparable to those of the reference networks under the considered multiphase flow scenarios. Conditional generation satisfies 99.6% of prescribed well observations in 2D. In 3D, where vertical boreholes constrain full voxel columns, conditional realizations satisfy 84.2% of observed well voxels while maintaining fracture volume and geometry comparable to the reference ensemble. These results demonstrate the potential of latent diffusion models as compact probabilistic representations for fracture-network characterization, ensemble generation, and sparse-observation conditioning.

We show that it is possible for three people to win the GHZ game with just the standard real valued misty state formalism, but each party will need to have two balls instead of one. That is, there will be six balls involved, rather than three. In fact, as a general rule, instead of using complex numbers in a quantum computer you can always add a qubit to "tag" complex numbers. But for proofs of non-locality, like the GHZ game, we cannot have a single universal tagging qubit, because all three parties would need access to it and therefore they could potentially transfer information via it. That's where we need to replace the one ball with three that each only need to be accessed locally.

Large multi-dwelling unit (MDU), student-housing and hospitality networks rarely store resident pre-shared keys (PSKs) on the access point (AP). Instead, the per-resident keys of ePSK, MPSK, DPSK, iPSK or PPSK services are held in an external RADIUS server or cloud database synchronised with the property-management system. Our earlier work, Companion-BSSID-Assisted PSK Discovery (CAPD), brought WPA3-SAE to multi-PSK networks by learning the client's PSK on a WPA2-only companion BSSID and steering the client with IEEE 802.11v to a hidden WPA2/WPA3 primary BSSID of the same SSID. CAPD assumes that the AP can test every PSK itself. This paper removes that assumption. We propose RADIUS-Assisted Companion PSK Discovery (RA-CAPD). The AP forwards the ANonce, the raw EAPOL-Key message 2 (SNonce, MIC, RSNE), the SSID, the client MAC, the BSSID and the AP MAC to the external server in RADIUS vendor-specific attributes (VSAs). The server validates the MIC against all provisioned keys, identifies the matching PSK, and returns it encrypted under the RADIUS shared secret (or a per-AP key for cloud databases), together with the unit, room or user identity and the VLAN. The AP binds the client to this PSK and identity, synchronises the binding across all APs in the domain, and moves the client to the configured profile: 802.11v-capable clients via a BSS Transition Management request, and other clients by disconnection and reconnection. On that profile, SAE-capable clients complete WPA3-SAE and WPA2-only clients complete WPA2-PSK, both with the learned key and no further search, so the companion BSSID serves only for onboarding. We present the design, attribute dictionary, algorithms, complexity and security properties, together with laboratory validation on commercial access points.

Generating an image with a diffusion or flow-matching model typically involves integrating a velocity field for hundreds of sequential steps. That computation suits a GPU poorly and an analog neuromorphic substrate well, where one fast, low-power physical relaxation could replace the whole sequence. Two things are needed: a generative algorithm whose computation is a relaxation, and a way to train it without backpropagation, whose global gradient transport is hard to realise in analog. Energy Matching supplies the first, sampling by gradient-descent relaxation on a time-independent potential - but obtaining the gradient that descent follows is itself a backward pass, once per integration step. We prove that a predictive coding network with a single linear injection term added to its energy relaxes to a state whose visible layer carries that gradient inherently, with an error that vanishes for small injection strengths. An Energy Matching model trained by backpropagation can therefore be loaded into this network unchanged and sampled by relaxation alone, matching the original model's FID to within 0.04 at CIFAR-10 and ImageNet-32 scale. The network can also be trained without backpropagation: Equilibrium Propagation reproduces backpropagation's parameter updates to a cosine similarity of 0.9998, and training end to end with those updates reaches an FID of 3.64 on unconditional CIFAR-10 against 3.54 for backpropagation. To our knowledge this is the first CIFAR-10 generative model trained with Equilibrium Propagation, and the first at competitive quality by any backpropagation-free rule. This quality does not require group normalisation or attention: an architecture of only convolutions and pointwise nonlinearities, both native to an analog crossbar, reaches an FID of 3.51 when trained by backpropagation.

We prove the persistence of conormal analyticity through the $H^r$ norm, with $r \geq 4$, for the 3D incompressible Euler equations in the half-space $\Omega = \mathbb{R}_+^3$. We also establish a lower bound on the radius of conormal analyticity which depends on $\|u\|_{H^r}$ and initial data. In particular, persistence is established in both of the spaces $X_\tau^{\rm co}$ and $A_\tau^{\rm co}$, the former being based on the conormal operator $Z_k = \phi^k \partial_3^k$ and the latter being based on $L^k = (\phi \partial_3)^k$. Finally, we show that the norms $\|\cdot \|_{X_\tau^{\rm co}}$ and $\|\cdot \|_{A_\tau^{\rm co}}$ are equivalent up to a small loss in radius, the proof of which uses explicit identities relating the two conormal operators.

This paper is devoted to ideal approximation theory in $n$-exangulated categories. We define ideal balanced pairs and explore their properties and equivalent characterizations. We prove that any complete ideal cotorsion triple $(\mathcal{I},\mathcal{K},\mathcal{J})$ satisfying suitable assumptions yields an ideal balanced pair $(\mathcal{I}, \mathcal{J})$, and the converse holds under additional hypotheses. We further show that for an ideal balanced pair, it is equivalent to an additive subfunctor with enough projective morphisms and enough injective morphisms.

Future-work statements offer a record of the questions researchers choose to leave open. Using them to describe a field, however, requires separating thematic change from changes in document coverage, topic assignment, and scholarly writing. FutureLens examines these measurement issues in a recovered snapshot of 5,655 ACL papers published during 2012–2022, including 4,005 with supplied future-work reference text. We compare non-negative matrix factorization (NMF) and latent Dirichlet allocation (LDA), aggregate topic presence at the paper level, and examine normalized prevalence, exploratory trend statistics, assignment stability, and agreement between input-derived text and references. Across available seeds, NMF has higher mean normalized pointwise mutual information (0.1387 versus 0.1280) and higher pairwise assignment stability, while LDA has greater keyword diversity (0.8900 versus 0.7750). Seed-zero results show increasing prevalence of broad model-training and question-related vocabulary, alongside substantial citation and document-structure content. Mean input/reference topic-set overlap is 0.5191 for NMF and 0.4924 for LDA; temporal rank agreement varies widely and does not imply agreement in prevalence levels. These findings support a descriptive account of the available corpus, with explicit qualifications for missing references, incomplete runs, and unverified reference provenance. The contribution is an empirical measurement study of longitudinal scholarly signals, not a new topic-modeling algorithm or a forecasting system.

Publishers and funders increasingly ask authors to disclose how AI tools contributed to a piece of research. In computational work the repository is where an author or an auditor would look for the answer, and we ask what it can actually supply. We audit two single-investigator repositories by the same author in the same field, each carrying a different class of agent residue: a prescriptive trace of standing rule files and a narrative trace of a running agent scratchpad, in repositories of 466 and 103 commits. The two are legible in opposite ways. 85.6% of the lines ever committed to the prescriptive trace are visible at the current revision, and a standing rule states a conclusion and discards the episode behind it; the narrative trace records episodes and shows 21.8%. Counted by revision the gap is far smaller: 15 of the prescriptive trace's 41 revisions leave no visible line, against 18 of the narrative trace's 41. The Co-Authored-By trailer, the field an automated audit reads, is one of three AI trailer keys Project A carries and covers 66 of the 121 commits carrying any. In Project B it has low sensitivity against an explicit within-repository proxy: 39 commits revise a file maintained to hold an agent's working notes and 1 (2.6%) carries the trailer. Project A's repository-level rate is unstable in time too, stepping from 2.7% to 17.7% across a boundary we did not specify in advance. Repository channels disagree, and neither rate can be read as an involvement rate; we do not isolate the cause, and say so. We give a field-by-field account of which provenance claims such traces support, which they cannot, and which they answer confidently and wrongly, construct an auditable evidence ledger, and derive the requirements this places on any system meant to record research provenance as it happens.

A fermion bilinear acquiring a non-zero vacuum expectation value and generating a mass gap underlies the Peierls instability, BCS superconductivity, the Nambu–Jona-Lasinio mechanism, and the Sachdev-Ye-Kitaev model. These mechanisms are usually discussed separately, and their structural similarities have been noted only in passing. We argue here that the correct universal object is not the bosonic order parameter but the fermionic condensate, and we show that the four mechanisms fall into two universality classes distinguished by a single dimensionless parameter, $r = \hat m_\phi^2 / \hat y^2$, the ratio of the scalar mass to the Yukawa coupling. The classes are separated by a critical line $r_c(N_f)$ in the plane of $r$ and the fermion flavour number $N_f$; they correspond to two distinct infrared fixed points of the renormalisation-group flow and differ in their critical exponents. We derive the classification from a large-$N_f$ expansion followed by a functional renormalisation-group analysis using the Wetterich equation, which gives the finite-$N_f$ correction $\hat y_*^2(N_f) = 0.1125/N_f + 0.0023/N_f^2$, a $12.5\%$ enhancement over the leading-order prediction. The classification is predictive for systems whose class is not already known; for the organic conductor (TMTSF)$_2$PF$_6$ under pressure, we predict that both the spin-density-wave and superconducting phases lie in the fermion-bilinear class. We discuss the Standard Model Higgs as a borderline case in the scalar-primary class, and identify the composite-Higgs scenario as the particle-physics realisation of the fermion-bilinear class. A self-consistent Su–Schrieffer–Heeger calculation for polyyne chains realises the Peierls mechanism concretely, and we list five falsifiable predictions, two within reach of current techniques.

Whether large language models can assess the innovativeness of academic publications matters for literature reviews, research funding screening, and academic recommendations. We introduce the Scientific Research Taste Benchmark (SciTaste), a gold-standard benchmark covering six disciplines: physics, mathematics, chemistry, biology, medicine, and finance. SciTaste measures research taste, that is, the ability to judge a paper's scientific value and potential, of which innovativeness is one of seven dimensions. It integrates metadata from NASA ADS, OpenAlex, and arXiv, aligns papers to field-specific classification systems, and scores each paper with the seven-dimensional Academic Taste Index. Dimension weights come from a bi-objective machine-learning approach (five regressors, including random forests) and are shared across all six disciplines. The benchmark includes 300 selected papers (50 per discipline) and a backup pool of 5,700 papers (950 per field) covering 1996-2023. SciTaste supports four evaluation modes: pure_qa, native_search, our structured tool-calling agent DiM (9 bibliometric tools in ADS/OpenAlex), and dimagent_v3_native. Outputs are evaluated by two complementary schemes: correlation with dimension-level accuracy, and WAPE relative error. Across ten state-of-the-art LLMs, tool-based structured evaluation generally outperformed toolless and raw web search modes, though gains were model-dependent. Physics is easiest; chemistry, biology, and medicine remain more challenging. SciTaste supplies the value-judgment substrate auto-research systems need to screen, rank, and prioritize candidate ideas. Dataset, harness, and scoring code are available on Hugging Face at https://huggingface.co/datasets/skylenage/SciTaste-benchmark.

Kaniadakis entropy deforms the horizon entropy through a single parameter $K$, and the holographic dark energy built on it (KHDE) reduces to the standard model as $K\to0$. We ask whether current data can detect this deformation when spatial curvature is free and no cosmic-microwave-background calibration is assumed. Combining Pantheon+SH0ES supernovae, with the absolute magnitude fitted against the Cepheid calibrators, 32 cosmic-chronometer measurements of $H(z)$ and DESI DR2 baryon acoustic oscillations with a free sound horizon, we find that non-flat KHDE fits the data as well as flat $\Lambda$CDM ($\Delta\chi^2=-2.1$ for three additional parameters) but leaves $K$ unconstrained: the profile $\chi^2$ changes by only $0.47$ between $K=0$ and its best value, and the bound $K<0.67$ (95%) reflects the validity domain of the model rather than the data. Information criteria therefore prefer $\Lambda$CDM ($\Delta\mathrm{AIC}=+3.9$, $\Delta\mathrm{BIC}=+20.2$). The joint fit gives $\Omega_{k0}=0.012\pm0.045$, three times tighter than supernovae alone, $H_0=72.04\pm0.85\,\mathrm{km\,s^{-1}\,Mpc^{-1}}$ and $r_d=139.34\pm1.76\,\mathrm{Mpc}$; chronometers and DESI alone prefer $H_0=68.17\pm1.93$, $2.4\sigma$ below the supernova value. In the fitted model, acceleration begins at $z_t=0.66\pm0.03$; a model-independent neural-network reconstruction of $H(z)$ from the chronometers gives a present-day deceleration parameter $q_0=-0.46\pm0.14$, consistent with the model. We also identify three analysis requirements which, if ignored, make the deformation numerically invisible, underestimate the $H_0$ uncertainty fourfold, or spuriously reject the model.

Modified extensions of the standard cosmological models often introduce complex parameter degeneracies that severely impact sampling convergence of the parameter space while analyzing the model observationally. In this work, we present an accelerated Bayesian inference technique using a Heteroscedastic Deep Ensemble with Monte Carlo Dropout incorporating distance-gated local residual corrections to emulate complex likelihood surfaces and avoid extrapolation beyond the trained region. We demonstrate this methodology on a unified dark energy cosmological model equipped with Finsler geometry. The model is studied using the recent observational datasets, including Cosmic Chronometers (CC), Pantheon$+$SH$0$ES, and DESI DR$2$ BAO measurements. While the conventional sampling procedure struggles with slow mixing of parameter chains ($\tau_{max}=1578.3$), the neural network, specially trained densely in high likelihood regions, yields converged and highly resolved posterior distributions of the model parameters without significantly compromising the minimum chi-square value. We further discuss the results for physical inferences and also compare the present dark energy model with $\Lambda\text{CDM}$ statistically using AIC and BIC.

Representation autoencoders (RAE), created by pairing a frozen pretrained self-supervised learning (SSL) encoder with a trained decoder, have recently demonstrated state-of-the-art performance on image generation. Their use in the audio domain, however, remains limited: text-to-audio (TTA) models are typically trained in the latent space of dedicated autoencoders where the encoder and the decoder are trained jointly with reconstruction objectives. This paper presents a study on audio representation autoencoders (AuRA) and their applicability to general audio TTA. We train decoders for two pretrained SSL audio encoders and evaluate their reconstruction quality against autoencoders used by recent TTA systems. After obtaining decoders for the SSL representations, we also train rectified flow models in the SSL and baseline latent spaces to isolate the autoencoder's effect on generation. We find that generation in SSL spaces achieves comparable or superior performance to generation in the baseline spaces, despite differences in reconstruction performance. These results highlight that reconstruction quality alone is not a sufficient proxy for generative performance.

We investigated the field-, temperature-, and time-dependent magnetic moment of a nitrogen-doped type Ib diamond in darkness and under monochromatic light with a low laser intensity of about 160 $\mu$W cm$^{-2}$. In darkness, below approximately 25 K, the magnetic moment exhibits unconventional hysteresis loops that can be explained by the coexistence of superparamagnetism and superconductivity in certain regions of the sample. Under illumination, this hysteresis largely disappears, leaving mainly the superparamagnetic contribution. Temperature-dependent measurements reveal an irreversibility below 25 K in darkness, which vanishes under illumination. Similarly, at 0.5 T and below 25 K, the time-dependent magnetization shows flux-creep-like behavior that disappears when the sample is illuminated. All light-induced changes are fully reversible: the magnetic properties measured in darkness are restored after the sample is heated above roughly 50 K. Removing the superparamagnetic contribution from magnetization measurements up to $\pm 14$ T reveals clear characteristics of a granular superconductor, namely, a well-defined penetration field $H_p \approx 2.6$ T and a critical field $H_{c2} \approx 10.6$ T. Based on theoretical considerations, we attribute the observed phenomena to the combined effects of internal stress and local variations in nitrogen concentration within localized regions that can be regarded as grains. This interpretation is supported by Raman and birefringence spectroscopy measurements on the same sample.

ETC2 fixes the texture representation and bit rate, leaving the encoder to trade search time for distortion. We present a configurable CPU encoder in which rules or small neural networks allocate conventional ETC2 searches; these searches construct every encoded block. Configurations span fast-encoder output through broader mode and endpoint searches. On 64 color textures used to evaluate and select configurations, the fastest measured configurations reaching at least Google Etc2Comp's pooled REC709 PSNR at efforts 0-49.5 run 30.7-40.9 times faster. A high-quality configuration exceeds effort-100 PSNR with 64.1% less time. At equal search quotas, learned action ranking and second-search allocation jointly gain 0.124 dB over rule-based choices. At high quality, learning saves time at a small distortion cost. Outputs retain standard ETC2 storage and hardware decoding.

Code generation has become a core capability of large language models (LLMs), and code data is central to the pre-training process that develops it. As obtaining public code of sufficient quality becomes more difficult, further progress cannot rely on corpus expansion alone. Tiered data management instead applies increasingly targeted processing to smaller data pools, producing data states suited to different training objectives. Following this perspective, we present UltraData-Code as four connected data states spanning repository archival at L0, standardized natural code at L1, algorithmic selection at L2, and task-oriented synthesis at L3. L0 preserves repository data and provenance for reprocessing, while L1 applies scalable filtering, normalization, and near-deduplication for broad pre-training. L2 selects high-quality algorithmically relevant files using language-adaptive relevance signals and quality constraints, yielding approximately 400B tokens of UltraData-Code-L2 across 11 programming languages for focused continual pre-training. L3 converts selected implementations into structured programming exercises, yielding approximately 150B tokens of UltraData-Code-L3 across the same 11 languages for task-oriented mid-training or supervised fine-tuning (SFT). Under controlled 10B-token continual pre-training of a 1B model, L2 improves pass@1 over L1 by 7.80 percentage points (pp) on EvalPlus and 5.13pp on MultiPL-E. Replacing half of the L2 training tokens with L3 yields a further gain of 8.42pp and 8.07pp on the two benchmarks, respectively. Both gains increase when the training budget scales to 100B tokens. We have released our datasets and classifiers to support further research on LLMs.

Spatial intelligence, the capacity to represent and reason about physical space, underlies embodied perception and real-world visual understanding. Existing benchmarks probe it through spatial relations, perspective-taking, and mental modeling. What remains underexamined is whether a model can maintain a coherent scene representation across views, since pairwise comparison alone cannot reveal which view departs from the truth. We thus introduce Same Space or Not (SSoN), a benchmark of cross-view contradiction resolution comprising 1,000 questions from seven indoor scene sources and six categories of controlled 3D edits. Each question presents three views of an unchanged scene and one view of an edited variant. Models must identify the inconsistent view and explain the contradiction without a designated trusted reference or a marked target. Our pipeline iteratively translates human review into generation constraints. Automated checks and exhaustive human validation screen out both trivial visual shortcuts and questions lacking sufficient evidence for a unique answer. Humans achieve 83.3% accuracy, compared with 39.7% for the strongest vision-language models (VLMs), falling to 34.9% when a valid visual explanation is also required. We investigate prompting, camera information, and 3D reconstruction as interventions. On a 100-question intervention set with a 24% baseline, prompting and 3D reconstruction remain below 50% answer accuracy, except when ground-truth camera information is supplied, which reaches 61% answer accuracy but only 48% True solve accuracy. On a held-out 201-question test set, Qwen2.5-VL-7B achieves 21.89% accuracy, 33.33% after answer-only fine-tuning, and 41.79% after fine-tuning on both rationales and answers. These results identify cross-view contradiction resolution as a substantial weakness of current VLMs that can be partly addressed by targeted supervision.

Given a finite family of combinatorial objects with a total order, ranking asks for the position of a given object, and unranking asks for the object occupying a given position. Both provide algorithmic access to families too large to enumerate. Kagey gave a polynomial-time algorithm for several position-restricted families and left open the case of $\{-1,1\}$-avoiding permutations. We show that when $|S| = 2$, the graph of forbidden squares is $2$-regular, a disjoint union of cycles, and that fixing a prefix leaves only paths and untouched cycles. This yields polynomial-time prefix counting, ranking and unranking for every two-element set of offsets, covering $\{-1,1\}$. The same structure gives a short proof that the number of $\{-1,1\}$-avoiding permutations equals the $n$-th ménage number for odd $n$ and exceeds it by $2$ for even $n$, proving a conjecture recorded in the OEIS entry A354152.

We study a modification of symmetric teleparallel gravity in which the function $f(Q,T)$ is linear in the non-metricity scalar $Q$ and the trace $T$ of the energy-momentum tensor, combined with a Barrow holographic dark energy (BHDE) density built from the Granda-Oliveros infrared cutoff. The resulting background equations are solved in a modified version of hi\_class and constrained with Pantheon+ supernovae, 32 cosmic chronometer $H(z)$ points, DESI DR2 baryon acoustic oscillations, and Planck 2018 CMB distance priors, sampled with MontePython. Two versions of the model are considered, one with the neutrino sector held massless and one with the sum of neutrino masses $\Sigma m_\nu$ left free. Across five data combinations, $H_0$ falls between 67.5 and 72.3 km s$^{-1}$ Mpc$^{-1}$, closer to the Planck value when cosmic chronometers dominate the fit and closer to the SH0ES value when Pantheon+ is included; adding $\Sigma m_\nu$ as a free parameter raises $H_0$ in four of the five combinations, by amounts ranging from a few hundredths of a km s$^{-1}$ Mpc$^{-1}$ up to about 1.5 km s$^{-1}$ Mpc$^{-1}$, while the CMB-only combination shifts slightly the other way, and the fit uncertainties loosen throughout. The 95% upper bound on $\Sigma m_\nu$ ranges from 0.13 to 0.22 eV depending on the data combination, comparable to standard $\Lambda$CDM constraints. Comparison with $\Lambda$CDM through the Akaike and Bayesian Information Criteria gives mixed results: the extended model is favored once Pantheon+ is included, particularly with massive neutrinos, but is disfavored for CMB-only and CMB+BAO fits, where the improvement in $\chi^2$ does not offset the larger parameter count. The Barrow exponent $\Delta$ stays below 0.005 throughout, and the Granda-Oliveros parameters $\lambda$ and $\gamma$ remain near 1 and 0.5 across all combinations.

We investigate a particular case of the New Exponential Infrared $f(T)$ teleparallel gravity model with the same number of free parameters as the standard $\Lambda$CDM model. Remarkably, due to the two real branches of the Lambert function $W$, the model admits two distinct cosmological solutions, denoted by $f^{(I)}$CDM and $f^{(II)}$CDM. The former exhibits a late-time phantom crossing, whereas the latter predicts an effective dark-energy density evolving from negative to positive values. Using a Markov Chain Monte Carlo analysis of the Planck CMB, DESI BAO, and Type Ia supernova data, we find that the $f^{(II)}$CDM branch develops a bimodal posterior distribution in the primordial parameters $A_s$ and $\tau_{ reio}$ for the Planck-only dataset. We show that this bimodality originates from the tension between the low-$\ell$ TT and EE likelihoods, which becomes pronounced in models exhibiting a significantly enhanced late ISW contribution relative to the $\Lambda$CDM model. Moreover, the inclusion of background observations drives this branch toward cosmologically unacceptable parameter values, rendering it observationally unviable. The $f^{(I)}$CDM branch, on the other hand, remains observationally viable, yielding $H_0=73.43\pm0.69~\mathrm{km\,s^{-1}\,Mpc^{-1}}$ for the Planck-only dataset and $H_0=72.55\pm0.32~\mathrm{km\,s^{-1}\,Mpc^{-1}}$ after including DESI, substantially alleviating the Hubble tension. Although this branch provides a slightly better fit than $\Lambda$CDM for Planck-only data, the addition of background observations reverses this preference in favor of the standard $\Lambda$CDM model. These results show that the cosmological viability of this model critically depends on the choice of the Lambert $W$ branch, with only the $f^{(I)}$CDM solution remaining compatible with current cosmological observations.

Raman spectroscopy generates large volumes of unlabeled data that are difficult to exploit with conventional supervised learning. This work introduces a multi-domain self-supervised learning framework based on a masked autoencoder (MAE) for learning transferable Raman spectral representations from heterogeneous biological, chemical, forensic, and materials datasets. The proposed pretraining strategy incorporates physics-aware objectives by combining masked intensity reconstruction with a sparse Voigt-profile decoder that models spectral peak structure through differentiable peak centers, widths, and amplitudes. A denoising objective is further included to improve robustness to spectral corruption. The learned encoder is evaluated on substance classification, component identification, concentration estimation, and spectral denoising under linear probing, attention-based probing, end-to-end fine-tuning, and training from scratch. Experimental results show that self-supervised pretraining is most beneficial for challenging tasks with limited labeled data. Attention-based probing also reveals substantially more useful information in the frozen representation than linear probing, indicating that conventional linear evaluation may underestimate representation quality. In addition, the benefits of pretraining become more pronounced with increasing encoder capacity. Overall, the proposed framework demonstrates that physics-aware, multi-domain self-supervised pretraining can improve label efficiency, robustness, and transferability across diverse Raman spectroscopy tasks.

Baryon acoustic oscillation (BAO) measurements from DESI DR2 have strengthened indications of a time-evolving dark energy. In this paper, we investigate observational constraints on the generalized equation of state (GEoS) for dark energy, $w(a) = w_0 - w_\beta (a^{\beta}- 1)/\beta$, with latest data from UV and X-ray observations of high-redshift quasars, DESI DR2 BAO, Cosmic Microwave Background (CMB) data (including lensing) from Planck and the Atacama Cosmology Telescope (ACT), and Pantheon+ Type Ia Supernovae (SNe Ia). The full Quasar+BAO+CMB+SNe Ia data combination yields stringent constraints on the EoS parameters of $w_0=-0.840^{+0.057}_{-0.084}$, $w_\beta=-0.90^{+0.88}_{-0.23}$, and $\beta=2.2^{+2.4}_{-5.2}$. Specially, the $\beta$ parameter, which describe a broad range of dark energy evolution, is in perfect agreement with the Chevallier–Polarski–Linder model ($\beta=1$) – the most widely used two-parameter parametrizations of dynamical dark energy. We also compare GEoS with other dynamical dark-energy parametrizations including CPL, Barboza–Alcaniz (BA) and Jassal–Bagla–Padmanabhan (JBP), and report three key findings. Firstly, we find compelling evidence for dynamical dark energy: all four EoS parametrizations yield lower information criterion (IC) values than $\Lambda$CDM for the full data combination, indicating a preference for EoS parametrizations under the Akaike information criterion (AIC) and the variance-based deviance information criterion (DIC). Secondly, although the unified phenomenological framework GEoS performs better than JBP, the current data do not strongly discriminate among different EoS parametrizations. Thirdly, the EoS trajectories of GEoS, CPL, and BA exhibit qualitatively similar transitions from the phantom regime at higher redshifts to the quintessence regime at late times, while the JBP parametrization exhibits a nonmonotonic evolution.

AI-based 3D generation models such as Hunyuan 3D and Meshy can produce visually appealing 3D shapes from an image or text prompt.However, they typically focus on an object's external appearance, often producing incomplete or implausible interior geometry. Consequently, the generated shapes are not guaranteed to be functionally valid, limiting their direct use in downstream applications such as simulation and fabrication. In this paper, we present a computational approach to functionalize AI-generated 3D shapes through interior geometry construction. Our key observations are that a part's functional role defines its interior geometry, and such geometry is often smooth, relatively simple, and closely related to the part's external shape. Inspired by these observations, we introduce an interior-aware part graph to encode information necessary for constructing interior geometry within an AI-generated 3D shape, and build the graph via VLM-based shape segmentation and graph attribute inference. Guided by the graph, we model the interior geometry of each part as a generalized cylinder and infer its parameters from the part's external geometry through an optimization. The resulting 3D shapes are functionally valid while preserving their AI-generated external appearance. We demonstrate our approach on a variety of AI-generated 3D shapes, including vessels, containers, and storage furniture, and fabricate selected examples for functional validation.

Clinical multimodal models can learn enrollment shortcuts shared across modalities and fail on new cohorts. We model these data through a latent structural causal model: disease state and background factors generate physiology through shared mechanisms, while cohorts differ in selection and measurement. Holding acquisition fixed, we identify conditional disease evidence without observing disease-state labels or recovering the full causal model. In common Gaussian coordinates, cross-cohort state alignment and natural-parameter changes identify a minimal selection statistic; conditioning on it removes enrollment-dependent evidence and yields an identifiable reference posterior within the specified shift family. State-dependent interactions can carry evidence invisible to individual-modality posteriors, motivating joint, information-preserving representations. DISCERN combines an information-lattice encoder, a latent-state model, and an outcome head through joint training. Under reversed synthetic enrollment on known latent coordinates, conditioning raises state AUROC from 0.043 to 0.883 and from 0.265 to 0.825, while random conditioning does not. Exploratory cross-database mortality experiments between eICU and MIMIC-IV improve AUROC over matched empirical risk minimization by 0.036 and 0.035. Learned-coordinate diagnostics distinguish preservation of state information from identification of the coordinates required by the estimator.

Line-parallel language models advance multiple lines concurrently for substantial speedups. However, frameworks like the Line-Coupled Language Model (LCLM) enforce a fixed commitment schedule across lines. Because dependencies vary within text, a fixed schedule cannot hold back a line whose prerequisites remain uncomputed, forcing premature commitments and cascading reasoning errors. To address this limitation, we introduce the Lattice Language Model (LatLM), making commitment schedules an adaptive decoding-time choice. Text occupies a 2D lattice where eligible frontier tokens are scored each round; a policy commits confident candidates while deferring unready ones for extended context. One forward pass supervises entire sampled trajectories via 2D causal masking, while an append-only cache retains exact states. At 482M parameters, we establish two findings. (1) LatLM matches a data- and capacity-matched autoregressive model on held-out text when decoded sequentially, while its adaptive parallel decoder achieves near-autoregressive accuracy on independent-line tasks using 25–41% as many decoding rounds. (2) When cross-line dependencies cause LCLM's fixed schedule to fail, deferred commitment substantially improves accuracy while retaining parallelism where dependencies permit. On mixed dependency chains, LatLM achieves 96.5% accuracy versus LCLM's 78.5%, using fewer than half the decoding rounds of the autoregressive baseline.

Automatically generating high-performance kernels with LLM-based agents promises substantial gains in development efficiency, but two obstacles remain. First, while increasing multi-turn interaction with execution environments is crucial for refining generation quality, continuous context accumulation rapidly exceeds the model's effective context window. Second, training long-horizon tasks via GRPO inherently incurs sparse rewards due to its reliance on terminal outcome evaluation, severely hindering step-level credit assignment; To address these obstacles, we propose ATRL (Ascend Triton Reinforcement Learning), a post-training method that refines GRPO into a compaction-aware, segment-level objective for long-horizon kernel generation. We also design a lightweight agent harness providing cascaded compilation, correctness, and profiling feedback, alongside an Ascend-specific knowledge base injected during agentic RL training and evaluation to compensate for the base model's missing platform expertise. Evaluations on NPUKernelBench and KernelBench, both executed on Ascend hardware, show that ATRL consistently surpasses all baselines, including GRPO, in both correctness and speedup.

Collaborative inference serves as a critical methodology in edge intelligence. Existing collaborative inference methods operate by splitting the deep neural network (DNN) model between edge devices and cloud servers. However, these methods need to load different models for multiple split points when the communication or device changes, thus resulting in long switching times and substantial storage overhead on the edge side. To address this issue, we propose a Dynamic Split Network (DS-Net), which not only enables efficient split point switching but also enhances the collaborative speed and accuracy across various scenarios. Specifically, we design a dynamic switching framework that enables flexible selection among multiple split points within a single model, thereby reducing storage, memory, and switching overhead. To mitigate the feature distribution shifts and task-relevant information loss caused by dynamic splitting and high-ratio compression, we further propose a hierarchical compensation mechanism. Local alignment calibrates the decoded feature distribution at each split point, while task-specific global compensation restores task-relevant information at the global level. In addition, we incorporate sample-adaptive global compensation selection into split-point selection based on sample complexity and runtime resource conditions. Experimental results demonstrate that DS-Net reduces memory overhead by 74.6% and achieves millisecond-level switching latency. Moreover, under a 10 Mbps bandwidth constraint on the Nano platform, DS-Net outperforms state-of-the-art collaborative methods by achieving 1.2% higher accuracy on the COCO dataset, reducing latency by 42.5%, and improving throughput by a factor of 1.8$\times$.

Physics-informed neural networks (PINNs) are trained to minimize the residual of a prescribed equation, so a low training loss does not indicate whether a coefficient in that equation was specified incorrectly. This study evaluates what can be inferred about the encoded parameter from the trained network alone. When the residual is linear in the parameter, written as $A - \nu B$, the least-squares estimate $\hat{\nu} = A^{\top}B/B^{\top}B$ is obtained from a single evaluation of the frozen network, without retraining or reference data. Its accuracy depends on how closely the network satisfies its own equation, quantified by a normalized residual $\rho$. Recovery error is associated with $\rho$ (log-log correlation 0.72 across 36 convection-diffusion networks), and empirical exponents of 0.63 to 1.20 across three equations decompose into one half plus a measurable angular term. For Burgers' equation, a single-model test based on a block-bootstrap standard error flagged 3 of 300 correctly specified networks (observed false-alarm rate 1%), all 2% misspecifications at a practical budget, and two thirds of 1% misspecifications after convergence. For the self-similar blowup rate of the Córdoba-Córdoba-Fontelos equation, where the rate also enters the architecture, the estimator recovers about half of the misspecification, whereas $\rho$ increases approximately quadratically and distinguishes a 0.2% error. When the incorrect rate enters both the equation and the architecture, errors of up to 2% were not detected at the precision reached. At practical budgets, the raw loss can be lower for the misspecified model, a scaling effect that $\rho$ does not share.