Papers

Safety evaluations of large language models commonly summarize harmful-output behavior with attack success rate (ASR). Yet the same non-harmful outcome can arise for very different reasons. A model may recover a harmful task and refuse it, fail to recover the task, or respond to something else entirely. Distinguishing these cases becomes especially important under intent-obscuring prompts, where a low ASR does not reveal whether the evaluated task was actually engaged. To make this distinction explicit, we pair ASR with operative understanding rate (UR), which measures whether a response both identifies the evaluated task and treats it as the task to be answered. Across interfaces, this paired view reveals substantial variation hidden by ASR: similar ASR values can correspond to sharply different recovery rates. Controlled English reconstructions show that recovery consistently improves as compressed prompts become more explicit, whereas ASR does not follow the same pattern. A complementary contrast comes from FormalLogic, where high recovery can still coincide with frequent harmful assistance. Together, these results show that non-harmful outcomes are not equally informative about model safety, motivating the joint reporting of intent recovery and ASR in LLM safety evaluation. Code and experiment inputs are available at https://github.com/kevinjiang0121-cyber/IRIS.

Large Language Models (LLMs) increasingly ship with explicit "thinking modes", yet their counterpart, "no-thinking", has received far less attention. We study LLMs' no-thinking behavior along two axes. a. How to measure no-thinking? Prior work typically defines no-thinking through proxies such as a disabled thinking mode or the absence of long traces. These proxies are unreliable: disabled thinking modes may still emit reasoning, while long traces may contain filler rather than genuine inference. We instead normalize each response into a pre-answer trace and final answer, and evaluate it at three levels: (i) Empty-Thinking Rate for strict answer-only compliance; (ii) instruction-aware Question-Pre-answer Relevance for similarity between the question and pre-answer trace; and (iii) LLM-as-judge Explicit Inference Rate for visible explicit inference. Together, these metrics distinguish answer-only output, relevant but non-inferential text, and explicit inference. b. How does no-thinking vary across tasks and models? We evaluate six prompting interventions on six LLMs across Boolean, multiple-choice, and open-ended questions. We find that explicit no-think controls cannot reliably eliminate visible inference. Models instead exhibit "Thinking Inertia": explicit inference persists even under strict controls and becomes more prevalent as the answer space opens. Accuracy remains stable on Boolean and multiple-choice tasks, whereas open-ended tasks reveal a trade-off between answer-only compliance and task accuracy. Rewriting the same questions across answer spaces shows that supplying candidate answers makes answer-only responses easier to produce. These findings establish no-thinking as a non-trivial capability: stopping explicit reasoning cannot be assumed from model settings or instructions alone and deserves systematic evaluation alongside reasoning ability.

Reliable crop-row perception is essential for autonomous agricultural robots, but low-cost deployment is constrained by computation, memory, power, and inference time. This paper addresses crop-row detection and navigation-line extraction using UltraLight Luma, a compact encoder-decoder segmentation network with a progressive learning strategy and only 13.1k trainable parameters. The predicted crop-row masks are processed using a post-processing algorithm to estimate row alignment for downstream visual servoing. UltraLight Luma improves parameter efficiency over U-Net, YOLOv8, and YOLOv26 while maintaining reliable crop-row detection performance. The model required only 25.23 mJ per inference, demonstrating its suitability for resource-constrained agricultural robots.

Stellar granulation imprints line-dependent wavelength shifts and profile variations that limit the precision of radial velocity (RV) measurements used for exoplanet detection and characterization. Understanding how the sensitivity of individual spectral lines to granulation depends on stellar parameters is therefore essential for mitigating granulation-induced RV signals. Here, we investigate the effect of metallicity using MURaM 3D magnetohydrodynamic simulations of stellar granulation including a small-scale dynamo for metallicities $\mathrm{M/H}=\{-1.0,-0.5,0.0,0.3,0.5\}$. We synthesize spectra of neutral and ionized lines spanning a broad range of lower excitation potentials and transition probabilities using the MPS-ATLAS code, assuming local thermodynamic equilibrium. We quantify the sensitivity to granulation using the continuum-intensity-weighted standard deviations of the line centre-of-gravity wavelength and equivalent width, denoted by $\sigma_\lambda$ and $\sigma_W$, respectively. Both metrics increase systematically with metallicity at disc centre owing to the combined effects of more vigorous convection and metallicity-dependent changes in the line opacity. We find that the correlation between the solar ($\mathrm{M/H}=0.0$) and non-solar sensitivity metrics remains high from the disk center to the stellar limb. This indicates that the relative ordering of line sensitivities is robust against changes in viewing angle and metallicity. Consequently, line masks calibrated on the solar spectrum can be reliably applied to radial-velocity measurements of stars with non-solar metallicities, provided they have solar effective temperature and surface gravity.

Massive stars play a central role in stellar and galactic evolution as progenitors of supernovae, neutron stars, black holes, and gravitational wave sources. Despite their importance, the dominant formation mechanisms of massive binaries remain poorly constrained observationally. We aim to use a large, homogeneous sample of eclipsing binaries to observationally constrain binary formation across a wide range of primary masses and test predictions from various formation scenarios. We combine photometric time series data from TESS with distance measurements from Gaia DR3 and near infrared photometry from 2MASS to analyze an all-sky sample of eclipsing binaries. By accounting for observational and geometric selection effects, we derive bias corrected estimates of the binary fractions and orbital parameter distributions across different mass bins. We derive bias corrected binary fractions and bias corrected distributions of orbital period and radius-to-separation ratio for systems spanning primary masses from 1.1 Msun to 47 Msun. For primary masses up to 5 Msun, the observed distributions show close agreement with theoretical predictions for systems formed via disk fragmentation followed by migration. At higher masses, we observe an apparent deficit of very short period systems relative to theoretical expectations, although this regime is also expected to be affected by incompleteness. Overall, the observed properties exhibit a gradual transition across the full range of primary masses. Within the orbital period range accessible to our sample, the results do not require a sharp transition in binary properties between F- and O-type stars. The observed continuity is consistent with disk fragmentation followed by migration acting across a broad range of stellar masses, although the present data do not uniquely require this formation pathway and alternative channels may also contribute.

The genome holds the blueprint that governs the biological properties of the cell. Consequently, advancing our knowledge of genomic function is crucial both for a broader understanding of biology and for continued biomedical advances. The success of large language models on natural language and protein sequences has motivated similar efforts on genomic data. However, standard genomic language models (gLMs) often require extremely large computational resources and still fall behind traditional methods on some downstream tasks. Recently, MSA-based pretraining has been proposed as an efficient alternative, but existing models are limited to short input contexts, restricting their use to short-range tasks, such as variant effect prediction. In this work, we present RAGenome, the first retrieval-based gLM that scales pretraining to longer contexts (100$\times$ longer than existing MSA-based gLMs), allowing it to capture both across-species evolutionary relationships and within-species longer-range interactions. Trained on whole-genome alignments from 100 vertebrates, RAGenome substantially improves the long-range capabilities of MSA-based gLMs, raising gene finding performance from 0.45 to 0.60, while remaining competitive on purely evolutionary-based tasks like prioritizing pathogenic variants. RAGenome provides competitive gLM performance at a fraction of the training cost, unifying evolutionary modeling and long-range capabilities within a single, flexible, scalable framework. Code is available at https://github.com/PanosAntoniadis/RAGenome.

We study the multivariate convolution model with additive independent noise, multidimensional signal and noise, aiming to recover the signal's cumulative distribution function from contaminated observations. The noise is ordinary smooth with anisotropic regularity, and the signal belongs to an anisotropic Nikol'skii density class. We extend to the multivariate setting an approximate minimum \(L^1\)-distance estimator based on integrated kernel density estimation, and establish matching upper and lower bounds for the \(L^1\)-risk over anisotropic Nikol'skii classes. To our knowledge, this is the first minimax rate, not a sharpening of existing bounds. Although the lower-bound scheme is classical, the test-function construction is novel: it distinguishes no active component from at least one active component. The rate is governed by a censoring mechanism through the positive part, a noise filter in a harmonic-mean representation of the exponent. We further propose a fully data-driven, rate-adaptive procedure selecting an optimal bandwidth vector over the full Nikol'skii scale. On Nikol'skii product density classes, the \(L^1\)-distance between cumulative distribution functions and the \(1\)-Wasserstein distance share the same minimax rate, though not equivalent in general, via coordinate-wise decoupling of the \(1\)-Wasserstein cost on product measures. These findings reveal a link between the two distances, whose full understanding remains an open question.

Near-term quantum approaches to combinatorial optimization are limited by qubit counts, circuit fidelity, sampling cost, and the difficulty of encoding constraints, while machine learning is increasingly used to configure and control quantum optimization workflows. We call such workflows adaptive: decisions conventionally fixed in advance, from formulation and penalties to shot budgets, backends, and whether to invoke a quantum processor at all, are made by learned policies that respond to the instance, the progress of the solve, or the hardware. This review examines three paradigms, AI for quantum optimization, quantum for AI-driven optimization, and AI-quantum co-optimization, and organizes the literature by the decision being learned rather than by application. A structured search identified a 119-paper corpus, of which 67 papers form an evidence core coded in detail. The evidence is considerably stronger for AI-assisted quantum optimization than for the reverse direction: learning already reduces quantum evaluations, improves initialization, supports decomposition and penalty control, and mitigates noise, whereas evidence that quantum computation improves learned optimizers remains largely confined to small-scale simulation. Experimental controls are thin: of 52 studies in the three paradigms, 22 include no classical baseline, the quantum contribution is fully isolated in 10 of 24 where an ablation applies, and the median largest experiment uses 16 qubits. We introduce an M0-M5 evidence hierarchy, from simulation to practical quantum advantage, and find no broadly convincing result at the highest level. We argue that scaling is increasingly a systems problem: the question is not only whether a problem fits on a quantum processor, but how classical and quantum resources should be allocated across the optimization process.

Affine frequency division multiplexing (AFDM) has attracted significant attention owing to its excellent backwards compatibility, full diversity achievability, as well as strong resilience to high Doppler and various hardware impairments. Existing AFDM channel parameter estimation methods, however, either require large pilot overhead or rely on computationally intensive grid search methods. This paper proposes a grid-search-independent, low-overhead, impulse-pilot-based channel parameter estimation method for AFDM systems by exploiting its inherent shift properties. Specifically, we show that the effect of a Doppler shift on the impulse pilot is equivalent to a delay shift in the time domain with a phase rotation. Additionally, both delay and Doppler shifts lead to cyclic shifts in the affine domain. By utilizing these properties in both the time and affine domains, two linear equations are formed to estimate the integer delay and Doppler of each path. The fractional part of the Doppler is estimated from the Doppler-induced leakage around the impulse pilot by exploiting the AFDM input-output relationship. For multipath channels, a successive interference cancellation (SIC) framework is developed, in which the strongest propagation path is first estimated, reconstructed, and canceled before the remaining paths are successively estimated. Simulation results demonstrate that the proposed method significantly reduces computational complexity and allows lower pilot overhead than the benchmarks. The proposed method also achieves improved channel parameter estimation and

For a $t$-graph $F$ and an integer $r\ge t$, the \emph{$r$-augmentation} $F(r)$ is the $r$-graph obtained by adding the same set of $r-t$ new vertices to every edge of $F$. We investigate the behaviour of the \emph{Turán density $\pi(F(r))$}, which is the asymptotically maximum edge density of a large $F(r)$-free $r$-graph, as a function of $r\to\infty$. Let $H_k^r$ be the $r$-augmentation of the complete $(k-1)$-graph on $k$ vertices; equivalently, $H_k^r$ is the (unique up to isomorphism) $r$-graph with $r+1$ vertices and $k$ edges. This paper determines the Turán density of each $r$-graph $H_k^r$ within a constant factor and estimates the lower bounds on $\pi(H_k^r)$ coming from the circular construction of Sidorenko. In particular, it is shown that if $k/\log r\to\infty$ then $\pi(H_k^r)=(1+o(1))k/r$. Also, we consider augmentations of graphs (that is, the case $t=2$) and prove that $\pi(F(r))\ge(\log r)^{-1-o(1)}/r$ for every 3-chromatic graph $F$, which comes close to the known upper bound $1/r$. Note that, for all other graphs $F$, the function $\pi(F(r))$ is known to be either identically 0 or uniformly bounded away from 0.

Autoregressive video diffusion enables causal video streaming without a bidirectional pass over the full clip, but existing few-step systems usually retain only the opening and most recent frames in a fixed-size KV cache. Once an event leaves this window, later frames can no longer attend to it, a failure we term mid-horizon forgetting. We present Memory Forcing, a few-step streaming method that preserves this missing history without increasing the cache size. Its Archive & Working Banks partition the cache into sink, archive, and working regions, retaining diverse intermediate events alongside recent motion under fixed memory. Because absolute temporal indices drift outside the training range, Bank-aware RoPE reassigns indices at attention time so each bank remains distinguishable. At 1.3B, Memory Forcing leads on longer clips, shows the smallest drop from 5s to 60s among methods reporting all four lengths, and preserves subjects and scenes through leave-and-return. The same design scales to Wan2.2 5B, producing more physically plausible, realistic, and dynamic videos and, to our knowledge, the first public 5B model on this forcing line.

Future 6G networks aim to expose sensing as a native service by reusing communication infrastructure. We study respiration sensing on a cell-free massive multiple-input multiple-output (MIMO) base station, where a 64-antenna channel must be fused into a breathing waveform. The state-of-the-art hand-crafted fusion is near-optimal in benign conditions. It collapses, however, under strong in-band motion interference, whose frequency falls inside the respiration band. We show that a learned complex-weight beamformer recovers respiration by spatial nulling, and that the remaining gap to a per-recording oracle can be closed at deployment by label-free test-time adaptation. Crucially, we identify which label-free signal makes this work. Frequency- and variance-based criteria cannot separate an in-band interferer from breathing. Our spatial-redundancy test-time adaptation (SR-TTA), which maximizes consistency across random antenna subsets under an out-of-band spectral veto, preserves benign performance in our tests. The respiration-rate error drops from 5.8 to 0.8 breaths per minute (bpm) under simulated in-band interference, and the pipeline maps onto the Open Radio Access Network (O-RAN) architecture as O-RAN distributed-unit (O-DU) range-gating, an adaptation xApp, and a calibration rApp. On real testbed recordings, a one-time cross-subject calibration plus SR-TTA reduces failures from 47% to 7%, drawing level with the hand-crafted combiner using label-free test-time adaptation.

Some defects in an AI-generated study can be identified from its artifacts; others require knowledge of what was approved before execution. We propose study contracts that bind declared experimental choices, run obligations and claim scope to recorded execution evidence, and distinguish this contract-relative verification from scientific truth. A diagnostic using eight self-authored clean/mutated pairs illustrates the information boundary. A deterministic checker applying a registered, fault-specific rule to approved and executed objects detected all eight registered mutations. Across eighteen recorded judge aliases given individual metadata-filtered packages without pair context or the registry-selected fault label, 104 of 144 mutated evaluation cases received defect flags; the remaining cases comprised 32 abstentions and eight terminal failures, with no explicit clean decisions on mutated cases. Some packages retained approval and execution fields, including digests. The prompt instructed judges to abstain when evidence was insufficient. These results characterize a deliberately information-asymmetric development setting; they do not isolate the effect of authoritative information from differences in task specification and rule selection, and they are not comparative verifier quality or agent reward hacking. We identify full-information comparisons, legitimate-adaptation controls and closed-loop agent evaluations as necessary tests of whether contract checks improve useful compliant completion under optimization.

We answer the following question by Frittaion, Nemoto, and Rathjen positively: "Is [independence of premise] an admissible rule of $\textsf{CZF}$ or any other familiar constructive/intuitionistic set theory $T$?" To be more precise, we show that whenever $\textsf{IZF}$ (or $\textsf{CZF}$ with full separation) derives a statement of the form $\lnot\psi \to \exists y\ \varphi(y)$ where $y$ does not occur in $\psi$, then $\textsf{IZF}$ (or $\textsf{CZF}$ with full separation) also derives $\exists y\ (\lnot\psi \to \varphi(y))$. Our proof method is a variant of the famous Friedman-Dragalin $A$-translation adapted to the context of set theory in a hereditary manner.

Gaussian splatting provides an explicit and efficiently rasterizable scene representation for robot navigation. However, individual Gaussian primitives may not reliably represent obstacles as they are jointly optimized through alpha-composited rendering from a finite set of reconstruction views. We propose 2DGS-Planner, a path planner for ground robots that reads planning-relevant geometry from a 2D Gaussian splatting (2DGS) map through rasterization, rather than treating individual Gaussian primitives as obstacles. During offline roadmap construction, multi-view attribution converts rendered normal dispersion into structural scores for non-ground disks supported by the reconstruction views. These scores guide adaptive node sampling on the ground. Path-aligned orthographic queries validate candidate edges, while cylindrical queries estimate local clearance fields that are cached on the edges. During online planning, graph search initializes a route, and path refinement reuses the cached fields while accounting for the robot's dimensions and ground constraints. Experiments demonstrate improved roadmap connectivity, more accurate clearance estimation, and higher planning success compared with the tested baselines. These results support rasterization as an effective geometric query interface for planning directly on Gaussian maps. Code and data are available at https://2dgs-planner.github.io/

Co-denoising pretrained representations such as DINO can substantially improve the training speed and quality of flow matching models, but it introduces a second denoising trajectory and requires carefully designed schedules. We propose a simpler alternative: predict the pretrained representation directly, then condition the model on its own prediction. This removes the need for a second ODE and any representation-specific denoising schedules, while retaining the benefits of representation guidance. Our approach converges substantially faster and achieves better generation quality as measured by FID score. On ImageNet, it outperforms the state of the art in latent space at 2x fewer epochs than prior methods; in pixel space, it improves FID over comparable prior methods by more than 20%. These results support a simple principle: do not denoise what you can predict. Our code is openly available at https://github.com/arijit-hub/dino_forcing.

Tree-based speculative decoding verifies multiple draft continuations in one target-model pass, but finite trees built from draft scores face a fundamental draft-target mismatch. We ask whether better exact verification can increase acceptance on a fixed tree and how target feedback can improve the tree itself. Through a target-flow view, we identify a canonical exit law and prove that one plus target coverage sharply bounds the expected output-block length, including the bonus token, of any exact path verifier. All optimal verifiers share the same exit and bonus-token law, already attained by representative predraw-and-follow and sequential residual verifiers. This yields Tree Exit Verification (TEV), an exact, level-parallel procedure using one exit-node decision and one bonus-token decision. The exit law also identifies missing target probability, providing node-level feedback for Exit-Guided Draft-Tree Training (ExitTrain) on inference-time draft trees. Experiments across dialogue, code, and mathematical reasoning validate fixed-tree equivalence: ExitTrain increases average output-block length by 13%, while TEV reduces verifier-stage latency by 15%, yielding a 14% end-to-end speedup over DDTree. Our results distinguish two opportunities: better draft trees for higher acceptance and more direct verification for lower latency. Code: https://github.com/hsj576/TEV.

Pulsars have narrow pulses, so their signals spread over many harmonics in a Fourier transform. Standard pulsar searches add the powers of those harmonics and discard their phases, which throws away the pulse shape and some of the signal. Folding the time series at each trial period keeps both, but is expensive even with the Fast Folding Algorithm (FFA). We describe CoherentSearch.jl, an open-source Julia code that reconstructs the pulse profile at every trial spin frequency directly from the Fourier transform. It interpolates the complex amplitudes of typically 60 harmonics, inverse transforms them into a profile, and tests that profile against boxcar templates of many widths. Because the input spectrum is normalized, the noise in every profile is known in closed form, so one threshold gives one false-alarm rate across the whole search. In an injection study of over 860,000 simulated pulsars at a matched false-alarm rate, the new search recovers 76% of the pulsars in white noise, against 64% for riptide's FFA in its recommended configuration and 42% for PRESTO's accelsearch, and its threshold does not move under red noise. On one CPU core it is 1.4-2.2 times faster than riptide at matched frequency coverage, it scales to 27 times faster on 48 cores, and modern GPUs run it up to nine times faster than a 20-core workstation.

U-Net inference for brain-tumor segmentation requires billions of multiply-accumulate operations, motivating hardware that can reduce computation dynamically rather than relying only on fixed precision or static model compression. Most-significant-digit-first (MSDF) arithmetic exposes the leading digits of a result during computation, enabling output-dependent decisions before the full value is generated. This paper presents an MSDF accelerator for quantized U-Net segmentation with a two-stage grouped processing element supporting signed INT8 operands and in-stream bias accumulation. Four runtime mechanisms operate directly on the output digit stream: exact early negative detection (END) in ReLU layers, exact sign-only decision making in the segmentation head, calibrated low-order-digit skipping, and calibrated pruning. The two approximate mechanisms are selected offline under an accuracy constraint, while execution requires only lightweight control and does not modify the stored weights. On a residual U-Net trained with nnU-Net for BraTS, the proposed mechanisms reduce digit cycles by 38.38% while achieving a mean Dice score of 80.58% on 73 held-out cases, compared with 81.20% for the floating-point model; the exact mechanisms alone reduce cycles by 18.79% without altering the quantized output. Synthesized in 45~nm, the processing element operates at 500~MHz, occupies 0.858~mm$^2$, and consumes 0.726~mJ per $192\times192$ patch under switching-activity-annotated power analysis. A projected eight-output accelerator with shared activation delivery achieves 16.6~ms latency and 1.67~mJ per patch.

We propose and analyze two second-order basis-update & Galerkin (BUG) time integration methods for dynamical low-rank approximation on Tucker tensors and tree tensor networks. Both are built from a sequential pre-computation sweep, after which the differential equations for all basis matrices and connecting tensors are evolved by a Galerkin method. Further, the proposed methods are rank-adaptive by construction. The first, the second-order parallel BUG integrator, solves all these differential equations fully in parallel, which is favorable on parallel architectures, followed by a sequential augmentation and truncation step. The second, the second-order augmented BUG integrator, gives up full parallelism but conserves norm and energy for Schrödinger equations and dissipates energy for gradient flows up to the truncation tolerance. For both integrators, we prove a second-order error bound that is robust with respect to small singular values of the matricization of connecting tensors. Numerical experiments for radiative transfer and quantum spin systems validate the theoretical findings.

Real-time video super-resolution requires high spatio-temporal fidelity under strict latency constraints, challenging diffusion models due to their iterative sampling cost and limited temporal coordination. We propose a streaming-aware framework that adapts pretrained single-image latent diffusion models for efficient video super-resolution (VSR) by exploiting the sequential structure of video streams. Our Cross-Step Attention mechanism reuses intermediate denoising features across adjacent frames and diffusion steps, enabling temporal information exchange without explicit temporal modeling. We further introduce Trajectory-Coupled Diffusion Scheduling, which aligns adjacent diffusion states and provides cleaner intermediate representations for cross-step conditioning, improving temporal coherence. These components are integrated into a streaming inference pipeline that incrementally propagates latent states across frames, reducing the effective computational complexity from $O(N \cdot S)$ to $O(N + S)$ for $N$ frames and $S$ diffusion steps. Experiments on REDS4 and YouHQ40-Test demonstrate improved perceptual quality and temporal realism while maintaining frame-wise stability. Our method achieves over 40 FPS at $512 \times 512$ resolution after cold start, enabling real-time VSR without explicit temporal modeling.

We compute the chiral-odd form factors $H_T$, $E_T$ and $\hT$ of the nucleon in light-cone QCD sum rules, separately for the $u$ and $d$ quarks, and organise them into the tensor monopole, dipole and quadrupole moments $g_T$, $\kappa_T$ and $Q_T$. The quadrupole moment, with no chiral-even counterpart at leading twist, and the isoscalar channel, whose sum rules were derived but never evaluated, are new in this framework. The form factors are read from a Lorentz basis independent only after canonical ordering of the Dirac strings; three of the eight surviving structures give the three form factors separately, and the one usually used for the tensor charge is not among the eight. Two exact results follow analytically. The $d$-quark contributions to $E_T$ and $\hT$ are equal and opposite in the chiral limit, broken in proportion to the quark mass and the twist-six amplitude $\mathcal{V}_6$, so the $d$-quark sector carries a single independent function; and the isoscalar tensor charge receives no leading-twist contribution, its twist-three terms cancelling between the flavours. Neither is visible without resolving the flavours. At $\mu^2=1$~GeV$^2$ the two distribution-amplitude sets give $g_T^{u-d}=1.27(7)$, $1.08(6)$, $\kappa_T^{u-d}=1.35(62)$, $1.58(59)$, $Q_T^{u-d}=-4.40(83)$, $-4.20(73)$ and $g_T^{u+d}=0.39(7)$, $0.38(6)$, $\kappa_T^{u+d}=4.47(62)$, $4.32(59)$, $Q_T^{u+d}=2.41(53)$, $1.56(30)$. The mean-field relation $2\hT^{u-d}=-E_T^{u-d}$, tested without any large-$N_c$ assumption, holds in sign and order of magnitude, with ratio $0.64(16)$ and $0.67(15)$ against the predicted unity. In the impact-parameter plane the moments displace the two flavour distributions in opposite transverse directions by nearly equal amounts, $\langle b_y\rangle=0.37$ and $-0.37$~fm, and elongate both across the polarisation axis, the $d$ quark some five to six times more strongly

Neural representations can encode more features than they have dimensions, a phenomenon known as superposition. We study the dimension needed to compute Boolean gates from such representations. For a single threshold layer with a Gaussian random dictionary and uniformly random sparse Boolean inputs, we derive sharp dimension thresholds under two error criteria. A vanishing expected error count can require more dimensions than correctness of every output with high probability. Shared reads explain the gap: rare realizations can produce many errors at once. The expected-count threshold has ball geometry, while joint reliability has box geometry when a gate is evaluated on every feature tuple. Optimizing shared readout weights and biases gives explicit thresholds for conjunction, disjunction, and majority. For pairwise conjunction, the analysis also describes the transition near the threshold, in agreement with exact simulations.

We present TraceRelay, an attention-aligned recurrent architecture that distributes persistent representations over a rolling sequence of low-dimensional traces. Local right looking attention forms increments from lower-layer representations; delivery is delayed until all attended inputs are in the causal past. A fixed additive phase recurrence accumulates the delayed increments, and left-looking attention reads the resulting residual augmented stream. A stride-wise prefix sum supports parallel prefill and bounded-buffer continuation. We study 36 small-model runs on Equal Repeats, bounded Dyck closing-type prediction, and causal Most-Freq generation, using three seeds per setting. At trained length 256, Equal Repeats models with recurrent phase inheritance reach 98.81-99.69% accuracy versus 50.73-51.63% for separately trained variants without inheritance, despite the latter receiving more updates. Accuracy drops sharply at lengths beyond the training range. At the longest evaluated lengths, models with more dimensions in the middle layer's recurrent traces perform better on Dyck (76.34% versus 55.49% close accuracy at length 4096), whereas models with fewer trace dimensions perform better on five-symbol Most-Freq (70.74% versus 55.60% exact generation at length 1024). These contrasting cases motivate further study of how the size of recurrent representations should be chosen for different tasks, without establishing a general rule across tasks or model configurations.

Machine-learning models trained directly on electronic spectra can predict spin–orbit-driven properties, yet models with comparable predictive accuracy may learn different electronic dependencies. We examine this problem using magnetic anisotropy in atomic-scale magnets and a validation strategy designed to assess the physical relevance of learned spectral relationships. We compare a bidirectional gated recurrent unit and a one-dimensional convolutional neural network, both trained to predict magnetic anisotropy energy (MAE) from spin- and orbital-resolved scalar-relativistic densities of states (SR-DOS). Despite comparable in-domain (ID) accuracy, the architectures learn only partly overlapping spectral dependencies. Shapley additive explanations identify spectral features shared between the two architectures that can be associated with plausible spin–orbit-coupling pathways consistent with second-order perturbation theory (PT2). Adding PT2-derived MAE contributions to the model inputs has little effect on ID performance but can improve transfer beyond the training domain, with gains differing between architectures. Controlled spectral perturbations further demonstrate that similar attribution patterns do not imply the same functional dependence of the predicted MAE on spectral weight. The contrasting responses provide a functional context for architecture-dependent out-of-domain transfer. End-to-end learning can identify candidate electronic signatures of magnetic anisotropy. Their credible microscopic interpretation, however, rests on convergent evidence from predictive performance, model comparison, physical theory, controlled spectral interventions, and evaluation under distribution shift.

Recovering latent geometric structure from graph observations is a well-studied problem in statistical inference. Kapralov, Trevisan, and Wrzos-Kaminska (2026) introduced the problem of recovering the coordinate structure of the Boolean hypercube from a small random sample of its edges. More specifically, there are $n=2^d$ vertices, each corresponding to a distinct "feature" vector in $\{\pm 1\}^d$. Between each pair whose feature vectors are at Hamming distance one, an edge is observed independently with probability $p$. As long as the expected degree $pd$ is $\gtrsim \log d = \log \log n$, we give a polynomial-time algorithm that, given only the graph of observed edges, correctly recovers the entire feature vector of all but a vanishing fraction of vertices. This matches the information-theoretic guarantee of Kapralov, Trevisan, and Wrzos-Kaminska in polynomial rather than exponential time, resolving the algorithmic question left open by their work. Our algorithm combines a degree-$4$ sum-of-squares certificate for the structure of balanced near-minimum cuts with a rounding scheme originally developed for tensor decomposition by Ma, Shi, and Steurer (2016). The analysis relies on two novel ingredients: a sum-of-squares version of the Friedgut-Kalai-Naor theorem in Boolean Fourier analysis and a spectral concentration result for the observed subgraph of the hypercube. Finally, we provide a justification for why higher-degree sum-of-squares might be needed by showing a limitation of the basic SDP relaxation of this problem.

Neural networks are typically trained using first-order methods and back-propagation. It is unclear whether this approach is optimal for morphological layers whose weight Jacobians are sparse and whose resulting parameter gradients can be poor. In this work, we propose a novel weight update method for morphological neural networks inspired from the Multiplicative Weights Update (MWU) scheme. We view each morphological perceptron as an instance of the learning from experts' advice problem in logarithmic space, and use a correlation-based reward that favors inputs aligned with the desired output change, regardless of whether a strong gradient signal has reached their weight. We empirically evaluate our approach by training fully connected layers both as stand-alone models and as parts of larger transformer networks. Across nine benchmarks, correlational training yields improvements on eight, by up to 32.84 percentage points, while substantially reducing run-to-run variability.

Radio Resource Management (RRM) is a fundamental challenge in 6G wireless networks, particularly under dynamic user demands, inter-cell interference, and heterogeneous QoS constraints. Centralized optimization solutions are often infeasible in practice due to the lack of system-wide state information, dynamic conditions, and signaling delays, making distributed learning-based approaches attractive. However, conventional multi-agent reinforcement learning (MARL) struggles with scalability and constraint satisfaction in such highly dynamic environments. We introduce a scalable two-phase hybrid learning framework to address the RRM challenges where orthogonal-frequency division multiplexing (OFDM) domain knowledge is explicitly incorporated into the MARL pipeline. In our proposed two-phase learning framework, the first phase allocates a minimum resource to satisfy users' QoS requirements based on channel statistics, avoiding the inefficiencies of training policies under hard QoS constraints. Subsequently, a multiagent system is employed in the second phase to optimally allocate the remaining resources for system throughput maximization. By exploiting the structural interference information, we propose a scalable MARL algorithm which decomposes the original learning problem into smaller subproblems that can be handled independently, thereby avoiding exponential growth of the action space without compromising performance. Extensive simulations in realistic scenarios with 50MHz bandwidth and different numerologies show that our method significantly outperforms existing optimization and learning baselines, offering up to 40% throughput improvement, and 100% constraint satisfaction with minimal resource usage.

Reliable uncertainty quantification (UQ) is essential for deploying neural networks in scientific and high-stakes applications, but full Bayesian inference over the network parameters is computationally infeasible. We propose a low-rank generalized Laplace approximation for neural-network UQ based on a small number of data-informed curvature directions. Starting from a generalized Bayesian posterior defined through an empirical loss, we construct a local Gaussian approximation around a pretrained set of weights in this active curvature subspace. The posterior variances in the retained subspace are available in closed form, and the prior variance is calibrated by an empirical Bayes procedure. The generalized Bayesian formulation allows us to compare two posterior scalings: the standard Bayesian scaling associated with the summed negative log likelihood, and a mean-loss scaling in which the empirical loss is normalized by the number of data. A central finding is that the standard scaling induces a data-size dependent contraction of the posterior variance in the leading active directions. In regression problems, this can force the low-rank framework to retain additional weak-curvature directions in order to achieve nominal coverage of calibration data. When posterior samples are propagated through the non-linear network, these additional directions can degrade the coherence of the predictive intervals and shift the posterior predictive mean away from the pretrained model. In contrast, the generalized mean-loss scaling yields a more stable, lower dimensional active subspace and produces calibrated, coherent predictive confidence intervals. These results indicate that generalized Laplace active subspaces provide a practical and scalable route to calibrated uncertainty quantification in neural networks.

Bayesian physics-informed neural networks (B-PINNs) are a popular framework for parameter and state inference from sparse or noisy observations. They are commonly formulated via a collider structure, in which physical and trajectory parameters are assumed to be a priori independent and become coupled through virtual likelihoods on differential-equation residuals that enforce physical consistency. We show that this modeling choice can induce severe systematic bias in the posterior over physical parameters: even when the prior is favorably centered on the ground-truth parameters, the resulting posterior can drift away and concentrate far from them. As a remedy, we advocate a hierarchical chain model in which physics generates trajectories, which in turn generate observations. The chain model does not suffer from this posterior bias, but it poses a harder, so-called doubly intractable, inference problem due to a physics-dependent normalization constant. This challenge can be resolved by discretizing the underlying stochastic dynamics, after which the chain posterior can be sampled exactly with particle MCMC. We identify two distinct mechanisms characterizing the collider bias, derive analytical approximations of their magnitudes, and establish diagnostic criteria for predicting when standard B-PINNs remain reliable. Experiments confirm the predicted bias and show that the chain formulation successfully avoids it.