Computer Vision
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Wednesday 9 September
Recent advances in Intrinsic Image Decomposition (IID) have increasingly relied on generative models. However, progress remains limited by three key challenges: (a) insufficient physical consistency, (b) high computational cost at inference time, and (c) limited generalization capabilities. In this work, we show that latent bridge matching (LBM) effectively addresses these limitations for albedo estimation. We introduce a novel LBM-based architecture that enforces physical consistency through a pixel reconstruction loss, benefits from the inherent efficiency of LBM low-cost inference, and improves generalization across diverse datasets by incorporating a shading conditioning. In this extended version, we additionally show that conditioning the shading estimator itself on the predicted albedo further improves reconstruction fidelity, and we benchmark our best model against stateof-the-art IID methods across five real and synthetic datasets.
Machine Learning
Forward-Free LLM Depth Pruning via Weight Redundancy
Depth pruning reduces large language model (LLM) inference cost by removing complete Transformer blocks. Activation-based methods collect hidden states through forward passes on calibration data, while existing forward-free methods score each Transformer block separately without measuring similarity between blocks. We propose Weight-Redundancy Pruning (WRP), a forward-free depth-pruning method that estimates inter-layer redundancy from checkpoint weights to select blocks without calibration data or model forward passes. WRP compares attention output and MLP down-projection weights across layers and combines their pairwise similarities with relative projection-scale information. The resulting all-pairs similarity matrix guides layer grouping and block selection. Across multiple pruning settings, model families, and downstream tasks, WRP consistently outperforms existing forward-free magnitude pruning and approaches the performance of activation-based methods.
Read the full abstract
Artificial Intelligence
Scored vs. Generated Readouts in Behavioral Language Models: An Empirical Study of Elicitation Format
Language models fine-tuned on customer behavior can predict outcomes and generate explanations, but these readouts are often treated as interchangeable. Holding model checkpoint and prompt content fixed, we compare probabilities obtained by scoring answer tokens with predictions generated after a written rationale. Across 13 model-domain cells covering four retail tasks in three markets, including two using fully public data and checkpoints, the scored readout ranks outcomes more accurately in 12 of 13 cells (two-sided sign test, p approximately 0.003), by 1.5 to 14.5 points in area under the receiver operating characteristic curve (AUC). Paired bootstrap confidence intervals exclude zero in every newly measured cell. The gap varies with task-specific supervision and mismatch between training and serving formats, ranging from -2.2 points for an untuned base model to +13.7 for rationale-format supervision. Analysis of approximately 9,000 rationales identifies two correlates: reduced reliance on the dominant predictive feature and convergence on stock formulations. Probability saturation does not track the gap. A third readout, eliciting a probability before any verdict, improves calibration (Brier score from 0.47 to 0.15) while ranking within noise of scoring, but only for outcome rates represented in training; it is worse than scoring when the scored head is already calibrated. We interpret these differences through the objectives matched by each readout, identify training choices that narrow the gap, and propose retaining generated rationales while sourcing ranking from the scored head.
Read the full abstract
Computer Vision
CLFTv2: Efficient Camera-LiDAR Fusion for Semantic Segmentation via Hierarchical Feature Pyramids
Semantic segmentation for autonomous driving requires reliable detection of vulnerable road users (VRUs) despite heavy class imbalance. We introduce CLFTv2, a hierarchical camera-LiDAR fusion framework replacing global ViT attention with a Swin-based multi-scale encoder and a lightweight FPN-style residual decoder. Operating in the 2D perspective domain, CLFTv2 integrates multi-scale geometric cues through shifted-window attention and per-scale residual fusion, avoiding the computational overhead of query-matching decoders. Across three driving datasets, CLFTv2 consistently improves VRU recall. On ZOD, CLFTv2-Large achieves 53.5% mIoU, improving pedestrian IoU from 35.5% to 44.9% over the prior CLFT model. On Waymo, CLFTv2 reaches 61.7% mIoU. Additionally, a modality-isolation study suggests ViT's global receptive field yields stronger fusion gains only under dense LiDAR returns. Compared to a Swin-based Mask2Former adaptation, CLFTv2 requires 1.4$\times$ fewer GFLOPs and delivers 2.2$\times$ higher throughput, while achieving comparable overall accuracy. These results demonstrate that hierarchical local-attention fusion offers an efficient, scalable alternative to global-attention and query-based decoders for real-time on-vehicle perception in intelligent transportation systems. Source code is publicly available.
Read the full abstract
Astrophysics of Galaxies
A Theoretical Study of the Stellar Parameters and Surface Nucleosynthesis of the Extremely Metal-Poor RGB Giant BD$-18^{\circ}$5550 - Part I: Effects of $\alpha$-enhancement and Internal Mixing
We present a detailed stellar evolution study of BD$-18^{\circ}$5550, an extremely metal-poor red giant star in the Galactic halo. Such stars are among the oldest known objects and preserve chemical signatures of the earliest stages of Galactic history. BD$-18^{\circ}$5550 exhibits $\alpha$-enhancement, $\rm[\alpha/Fe]=+0.4$, typical of Population II halo stars, with enhanced abundances of oxygen, magnesium, and other $\alpha$-elements relative to iron. Using the MESA stellar evolution code, we perform the first direct determination of its mass, age, radius, and luminosity by simultaneously fitting eight observed stellar properties and multiple chemical abundance ratios. Our analysis yields a current mass of $\rm M=0.76,M_{\odot}$, radius of $R=38.4,R_{\odot}$, luminosity of $\log L=2.8,L_{\odot}$, and an age of $t=11.87$ Gyr, with a 7-fold improvement in age precision. The model reproduces all observed constraints, supporting our treatment of stellar physics, nuclear reactions, and internal mixing. We further investigate the predicted surface chemical composition up to iron and compare it with observations. The surface abundances of elements up to iron remain essentially unchanged throughout the evolution, while nitrogen shows gradual enrichment, indicating the onset of mixing processes at intermediate RGB stages. The preservation of the other surface abundances provides a valuable window into the chemical conditions of the early Galactic environment. Our results establish BD$-18^{\circ}$5550 as a benchmark star for constraining stellar evolution and early Galactic chemical evolution. Its derived age places it among the oldest stars in the Milky Way, providing important constraints on the epoch of early star formation and the initial assembly and evolution of our Galaxy.
Read the full abstract
Rings and Algebras
Lifting strongly graded rings via overgroups
Motivated by lifting problems for principal bundles, we study the following algebraic problem. Let $\smash{\widehat{G}}$ be a group with identity element $e$, let $G \leq \smash{\widehat{G}}$, and let $S$ be a strongly $G$-graded unital ring with principal component $R:=S_e$. We ask whether the given grading extends to a strong $\smash{\widehat{G}}$-grading without altering its $G$-homogeneous components. For every extension $\smash{\widehat{p}}$ of the Picard homomorphism of $S$, we construct a characteristic class in the relative third cohomology group \[ H^3_{\smash{\widehat{p}}}(\smash{\widehat{G}},G;\operatorname{U}(Z(R))), \] where $Z(R)$ denotes the center of $R$ and $\operatorname{U}(Z(R))$ its group of units. The~vanishing of this class is equivalent to the existence of a lift. When it vanishes, the relative second cohomology group \[ H^2_{\smash{\widehat{p}}}(\smash{\widehat{G}},G;\operatorname{U}(Z(R))) \] acts simply transitively on the equivalence classes of lifts with Picard homomorphism $\smash{\widehat{p}}$. We also prove that an $H$-grading is strong if and only if both its restriction to a normal subgroup $N \trianglelefteq H$ and its induced $H/N$-grading are strong. We illustrate the theory through a range of examples.
Read the full abstract
Information Retrieval
Purchase Advice and Observable Buyer Responses in Real AI Conversations
How often does a generative assistant persuade someone to buy, or persuade them not to buy? Conversation logs contain recommendations, but they do not necessarily record subsequent decisions. We audit 317 historical interactions from Aiso's proprietary research database of licensed, consent-based, de-identified conversations with commercially available AI assistants. Single-agent AI-assisted screening identifies 68 purchase-directed records; collapsing one shared-prefix copy yields 67 retained episodes, dated April 2023 to July 2025. Assistant responses provide candidate options, acquisition channels, or conditional preferences in 52 episodes (77.6%). One episode contains conditional redirection away from a named accommodation candidate. No episode is coded as advice to abandon or defer the purchase category. Only 18 episodes (26.9%) contain a subsequent user turn within the same purchase-related mission, compared with 23 (34.3%) that contain any later user turn. Using conversation depth alone therefore overstates this follow-up availability by 27.8%. Across 47 retained user follow-up messages, no explicit post-advice purchase commitment, completed-purchase report, or purchase-category abandonment statement is observed. These zeros describe recorded statements, not conversion or persuasion rates. The paper supplies operational definitions, text-free annotations, and reproducible descriptive results. Its central finding is a measurement limitation: recommendation content is observable much more often than a buyer's subsequent decision. The selected historical sample, unvalidated AI annotations, and missing transaction outcomes do not support a population-level or causal estimate of persuasion.
Read the full abstract
Solar and Stellar Astrophysics
Studies of Two Post Common Envelope Binary Stars
New time series photometry of the short period ($P<0.2$~d) binary stars ZTF~J032906.36+070408.3 (ZTF~0329) and ATO~J090.7238+00.5996 (ATL~0602) is presented. The primary (hotter) stars in both systems are white dwarfs, while the cooler stars are M dwarfs. Archival photometry and new and archival spectra are used to derive the temperatures of the components in the two binaries. Both are single-lined binaries. Velocity curves are fitted to radial velocity measurements of the red dwarfs and used to place lower limits on the masses of the white dwarfs. New and archival time series photometry show evolution in the shapes of the light curves of both stars. Spectra of ZTF~0329 show an excess of radiation in the $R_C$ band which is ascribed to electron cyclotron radiation, suggesting that the star is a low accretion rate polar. Satisfactory model fits to the multi-filter light curves of ATL~0602 require the presence of cool spots on the red component of the binary. The latter star exhibits a variety of velocities associated with Balmer and Ca~II emission lines within individual spectra.
Read the full abstract
Computer Vision
From Pixels to Hierarchical Sequences: Quadtree Mask Encoding for Vision-Language Binary Change Detection
Dense change detection in remote sensing requires vision-language models (VLMs) to compare bi-temporal images and generate accurate pixel-level masks. Existing VLMs are largely confined to change captioning outputs, and the few that produce pixel-level masks still rely on external decoders or flat text-as-mask serialization, which are less effective for small and fragmented changes. We introduce QUAKE-CD, a framework that recasts dense change prediction as syntax-verifiable structured generation. QUAKE-CD represents binary change masks as grammar-constrained quadtree token sequences, making the masks compact, syntactically checkable, and deterministically decodable within an autoregressive generation space. We further construct QUAKE-CoT, which pairs these sequences with chain-of-thought traces grounded in visual evidence, and jointly optimizes textual reasoning and spatial dense prediction through a progressive curriculum followed by grammar-gated dual-reward RL. On QUAKE-CoT, QUAKE-CD achieves 78.31% accumulated F1, outperforming decoder-based and flat text-as-mask VLMs while producing more faithful bi-temporal reasoning.
Read the full abstract
Artificial Intelligence
AgentAudit: An Open, Extensible Framework for Full-Lifecycle Trust Evaluation of AI Agents
Existing evaluation frameworks mostly assess only one part of AI agents, such as task completion (AgentBench) or security robustness (AgentDojo, ASB), rather than the complete pipeline of planning, tool selection, tool execution, memory and reasoning. Failures can occur at any stage, yet existing benchmarks rarely identify their precise source. AgentAudit evaluates the entire execution trace across ten capability, grounding, security and behavioural dimensions, namely instruction integrity, planner, memory, tool selection, tool invocation, tool correctness, alignment, tool faithfulness, security and execution integrity, combined with behavioural classification and failure attribution to pinpoint the exact stage responsible for an observed failure. AgentAudit can evaluate any LLM-based AI agent, since it attaches to the agent instead of replacing it. It reads only the recorded execution trace and does not interfere with how the agent runs, so it places no constraint on the agent's internal implementation. We evaluate five language models (OpenAI GPT-5, Claude Sonnet 5, Sarvam 105B, Llama 3.3 70B and Gemini 2.5 Flash) across nine capability and adversarial tasks. Claude Sonnet 5 and GPT-5 obtain the highest mean Composite Trust Scores (95.1 and 80.6 out of 100, respectively), while Sarvam 105B, Llama 3.3 70B and Gemini 2.5 Flash trail substantially (57.6, 45.7 and 22.6). All traces were scored by a single fixed judge model, which was itself one of the evaluated models, a limitation discussed in Section VII.E. More importantly, models with similar task-completion behaviour can diverge sharply in trustworthiness, as several non-frontier models are repeatedly classified Unsafe_Compliance on adversarial tasks rather than merely failing them, a distinction that pass/fail benchmarks cannot surface.
Read the full abstract
Statistical Mechanics
Quantifying cell shape and density fluctuations in epithelial tissue in vivo
Controlling changes in cell shape are crucial for many biological processes, such as tissue development and wound healing. Tissues typically are heterogeneous with a variety of cell shapes and sizes. Of particular interest are local deviations from an average cell shape and size. These fluctuations may extend and transmit across tissues, potentially offering valuable insights into tissue characteristics such as variations in effective "stiffness" or rigidity. In this study, we present a theoretical framework that captures the dynamics of epithelial cell shapes within tissue, incorporating both their average behaviour and fluctuation patterns. We model cells as interacting soft ellipsoids of varying size and aspect ratio. Coarse-graining our model, we obtain a set of continuum stochastic differential equations from which we derive spatial-temporal correlation functions. These correlation functions fit with our experimental data from the developmental process of the Drosophila pupal wing. From the correlation functions, critical parameters representing active cell shape changes and effective tissue "stiffness" can be determined.
Read the full abstract
Combinatorics
Boundedness and Separation Between Induced and Non-Induced Covering Numbers
There are four covering numbers $\mathrm{c}_g^{\mathcal{G}}(H),\mathrm{c}_u^{\mathcal{G}}(H),\mathrm{c}_l^{\mathcal{G}}(H),\mathrm{c}_f^{\mathcal{G}}(H)$, each of which measures in a slightly different way how well the edges of a graph $H$ (called a host) can be covered with graphs of a class $\mathcal{G}$ (called a guest class). If we require the graphs of $\mathcal{G}$ to correspond to induced subgraphs of $H$, we obtain an induced variant $\mathrm{ic}_x^{\mathcal{G}}$ for each covering number $\mathrm{c}_x^{\mathcal{G}}$ which satisfies $\mathrm{c}_x^{\mathcal{G}}(H) \leq \mathrm{ic}_x^{\mathcal{G}}(H)$ for every graph $H$. Yet, in general $\mathrm{ic}_x^{\mathcal{G}}$ cannot be bounded in terms of $\mathrm{c}_x^{\mathcal{G}}$. If there exists for a guest class $\mathcal{G}$ and a host class $\mathcal{H}$ a function $f$ such that $\mathrm{ic}_x^{\mathcal{G}}(H) \leq f(\mathrm{c}_x^{\mathcal{G}}(H))$ for every graph $H \in \mathcal{H}$, we call $f$ a binding function. Within this work, we study for which structural properties of a guest class $\mathcal{G}$ and a host class $\mathcal{H}$ such binding functions exist. We consider guest classes $\mathcal{G}$ that are monotone, hereditary, component-closed or neither, and have bounded maximum average degree, bounded chromatic number or neither. The host classes $\mathcal{H}$ we consider have bounded treewidth, exclude some minor, have bounded maximum average degree, bounded chromatic number, or none of these properties. For $219$ out of the $240$ possible $3$-tuples of properties for $\mathcal{G}$ and $\mathcal{H}$ and covering numbers we either provide a binding function or an example where no such function exists. In particular, we show that such binding functions always exist for hereditary guest classes $\mathcal{G}$ of bounded maximum average degree for three of the four covering numbers, but may not for the fourth kind.
Read the full abstract
Numerical Analysis
Initialization-dependent BFGS trial rates for tilted absolute values
Lewis and Overton conjectured in 2008 that, for BFGS applied to the one-dimensional tilted absolute value with their Armijo–Wolfe line search, every nonterminating execution converges at a trial-normalized rate determined only by the tilt parameter and independent of the initial data. We prove that this initialization-independent rate assertion is false. In fact, for every tilt parameter in an explicit open interval, we construct two nonterminating executions with different sharp trial-normalized convergence rates. The proof reduces the line search to a scale-free state consisting of the iterate sign and the zero-crossing stepsize, and verifies two periodic state cycles by elementary rational inequalities. This yields an explicit family of counterexamples and shows that the asymptotic trial-normalized behavior of nonsmooth BFGS can depend essentially on the initialization even for this canonical one-dimensional model.
Read the full abstract
K-Theory and Homology
The Oka principle for étale Chow groups
The celebrated theorems of Shilov, Arens–Royden, and Forster give direct descriptions of the first three integral cohomology groups of the Gelfand spectrum of a commutative complex Banach algebra. In his 1974 ICM address, Taylor asked whether the higher cohomology groups admit descriptions in terms of the underlying ring. We give a solution to this question in even degrees: The étale (aka Lichtenbaum) Chow group in every codimension is canonically isomorphic to the corresponding even integral cohomology group of the Gelfand spectrum.
Read the full abstract
Mesoscale and Nanoscale Physics
Photo-induced Wavelength-tuning of Telecom-band Quantum Dot Nanowires Embedded in a Phase Change Material
Scalable quantum-photonic technologies require spatially separated emitters to emit at the same wavelength, yet epitaxially grown quantum dots naturally exhibit emitter-to-emitter variations in their emission energies. Here, we demonstrate a contact-free post-growth tuning approach for InAs quantum dots embedded in InP nanowires using the photo-induced transformation of an amorphous Sb2S3 shell. This photoinduced effect leads to a progressive relaxation of the strain imposed on the InP core, enabling controlled spectral tuning of the quantum-dot emission. The effect is observed both at room temperature and under cryogenic conditions for quantum dots emitting in the O-band of the telecommunications spectrum. Strain-induced redshifts of approximately 7 meV at room temperature and 28 meV under cryogenic conditions are observed, respectively. The use of an amorphous chalcogenide shell therefore provides a route to achieve local wavelength tuning at the single quantum-dot-in-a-nanowire level after growth. This approach could enable spectrally matched emitters for scalable quantum-photonic architectures.
Read the full abstract
Algebraic Geometry
Positivity of compact Kähler varieties admitting an int-amplified endomorphism
We study compact Kähler varieties admitting int-amplified endomorphisms from the viewpoint of positivity of tangent sheaves. Our main result shows that the tangent sheaf of such a variety is weakly positively curved; in particular, it is pseudo-effective in a strong sense. This establishes a new link between complex dynamics and the positivity theory of tangent sheaves, and provides an alternative to equivariant MMP techniques in the compact Kähler setting. As an application, we prove a Kähler analogue of Yoshikawa's structure theorem: for a compact Kähler klt variety admitting an int-amplified endomorphism, after an equivariant quasi-étale cover, it admits an equivariant flat MRC fibration with irreducible fibres onto a complex torus; in the smooth case, the fibration is smooth whose periodic fibres are of Fano type. To this end, we prove the existence of minimal models in the case of numerical dimension zero canonical divisors for non-projective Kähler varieties. We further study rationally connected manifolds with pseudo-effective tangent bundle, focusing on their relation to Fano-type properties, almost homogeneity, and fundamental groups. In the proof, we also show that every surjective endomorphism of a compact klt Kähler variety lifts to a suitable maximally quasi-étale cover.
Read the full abstract
Systems and Control
Optimal Dispatch of a Hydrogen-Colocated Renewable-Powered Desalination Plant
This paper develops an analytical framework for profit-maximizing dispatch of water, electricity, and green hydrogen in a renewable-powered water desalination plant (WDP) combining thermal and reverse osmosis (RO) desalination. The optimal dispatch reveals that the schedules of the desalination units, electrolyzer, and grid interaction can all be characterized in closed form as functions of renewable generation. At low renewable output, the plant relies on thermal desalination and grid power while maintaining RO desalination and hydrogen production at their minimum setpoints. At high output, it reduces thermal desalination, increases RO desalination, and allocates surplus power and water to hydrogen production. For renewable output within a precomputed intermediate range, the WDP balances energy internally, allocating resources based on the relative marginal values of hydrogen and RO water. Simulations using real solar and plant data show that integrating RO and thermal desalination with renewable generation and hydrogen production achieves the highest daily profit of 75.945 k$, exceeding the best configuration without hydrogen by 15.1%.
Read the full abstract
Databases
Contextual Utility of Quantization Moves in Extreme Low-Bit LLMs
Accurate measurements of isolated weight replacements need not predict their joint effect on a quantized model. We study this gap for legal replacements at a fixed bitrate, separating isolated pricing error, composition error, and changes in evaluation data. On Llama-3.2-1B and activation-ordered 3B GPTQ bases, option-KL composition residuals are 26x and 42x the sums of absolute midpoint pricing errors; large residuals remain with exact-singleton prices. Exhaustive Qwen3-0.6B lattices reveal context-dependent effects: six of eight moves in the stratified lattice have statistically resolved ARC-KL sign reversals, excluding an exact monotone additive representation. Pair measurements improve prediction of unqueried combinations on both Qwen and Llama. Their value for selection depends on the criterion: the stratified Qwen front retains a quality-coverage advantage after refitting, whereas the 3B additive and quadratic fronts are nearly equal in quality at a one-standard-error tolerance. Low-order endpoint structure leaves continuous gradient responses undetermined, and shared-midpoint prices can differ from those of the executed subset. Support-matched controls show that restricting the number of replacements curbs the failures of large combinations, while additional endpoint search gives bank-dependent NLL/ARC trade-offs.
Read the full abstract
Audio and Speech Processing
UniStream: Multi-Expert Residual Vector Quantization for 48 kHz Causal Streaming Audio Coding
We present UniStream, a fully causal 48 kHz neural audio codec for streaming speech, music, and environmental sounds. At its core is Multi-Expert Residual Vector Quantization (ME-RVQ), which replaces the single shared codebook in each residual quantization layer with four expert codebooks controlled by a deterministic Top-K router. Because routing decisions are derived solely from previously decoded quantized states, the decoder can reproduce the selected experts without transmitting expert identifiers, thereby expanding quantization capacity while adding 5.5M parameters. We further introduce an auxiliary Optimal Transport Conditional Flow Matching (OT-CFM) objective to regularize the quantized latent space during training. The flow module is removed entirely at inference and therefore incurs no runtime overhead. UniStream supports a 12 kbps Top-1 mode and a 22.5 kbps Top-2 mode within a causal 48 kHz encoder-decoder framework, while achieving real-time GPU inference. To complement narrow-band speech metrics, we report 48 kHz ViSQOL audio mode, ViSQOL speech mode, standard VGGish-FAD, DNSMOS P.835, and higher-rate reference comparisons with Opus and EnCodec. At 12 kbps, UniStream-Top1 achieves PESQ and UTMOS scores comparable to EnCodec while reducing speech Mel-D from 13.07 to 8.21. At 22.5 kbps, UniStream-Top2 achieves a ViSQOL speech-mode score of 4.67 and an environmental audio-mode score of 3.96, exceeding all evaluated systems operating at 12 kbps or below in the latter setting. It also comes within 0.03 MOS-LQO of Opus at 24 kbps on speech in ViSQOL audio mode. Ablation studies confirm that ME-RVQ is the primary source of quality improvement, whereas OT-CFM provides perceptual gains on speech with a mild trade-off in spectral distortion.
Read the full abstract
Software Engineering
Keep Evaluation Fair: Detecting Data Leakage in Code Generation Benchmarks via Membership Inference Attacks
Code generation benchmarks are widely used to evaluate Large Language Models (LLMs), but benchmark data leakage into training sets can inflate performance and undermine evaluation validity. DetectLeak, a method specifically designed for code generation benchmark leakage detection, relies on perplexity scores to identify likely leaked samples. However, perplexity mainly reflects general familiarity with code patterns and may perform poorly on complex or rare samples. It also overlooks other useful signals, such as code similarity, functional correctness, and semantic representations. To address these limitations, we propose CGMIA (Code-Generation-specific Membership Inference Attack), a method for detecting leakage in code generation benchmarks. CGMIA fine-tunes a shadow model on a subset of benchmark samples to construct labeled member and non-member data. For each sample, it collects the input prompt, generated code, and reference solution, and extracts expert features, including CodeBLEU, edit distance, test pass rate, and perplexity, together with semantic features from CodeBERT embeddings. An integrated learning module combines these features to capture both surface-level memorization signals and deeper behavioral patterns, enabling a classifier to predict whether a sample was included in the target model's training set. Experiments on eight code generation benchmarks show that CGMIA outperforms eight existing membership inference methods in most cases. It also effectively detects known leaked APPS samples in StarCoder-7B's training data.
Read the full abstract
Artificial Intelligence
Shifting Relational Paradigms for Affective Computing: Affective Resonance, Vitality Affects, and Vocal Interaction Fields
Affective computing has largely followed an individual-state paradigm, extracting discrete emotion labels or arousal/valence from isolated speakers. We argue this framing is incomplete for interaction. Drawing on affective resonance and vitality-contour accounts, we propose a relational framework in which the primary unit of affective analysis is the interactional field constituted within vocal dynamics. As a proof of concept, we present a preliminary empirical study using continuous self-supervised speech representations to detect directional expressive coupling in multi-party conversation. Coupling is regime-specific, concentrated at sub-second timescales, and collapses under exclusive-speech negative controls, consistent with a relational account of affective dynamics. We introduce design frameworks for Artificial Affective Resonance Intelligence grounded in Affective Resonance Dynamic Ontologies, supported by null-calibrated directional coupling analyses across interaction regimes.
Read the full abstract
Computer Vision
Pretraining and Distillation Matter More Than Architecture Family for Label-Free Single-Cell Classification
Choosing a deep learning architecture for label-free single-cell classification remains an open question, with microscopy benchmarks reporting conflicting conclusions about CNNs versus transformers. We present a controlled benchmark on LIVECell phase-contrast microscopy data using source-image-disjoint train/validation/test splits to prevent parent-image leakage and matched optimisation, augmentation, and evaluation protocols across EfficientNet, Vision Transformer (ViT), and EVA-02 models. This allows the effects of architecture, pretraining, fine-tuning, tokenisation, and distillation to be disentangled. We find that the previously reported CNN advantage is largely explained by pretraining rather than architecture: the smallest pretrained model outperforms the strongest model trained from scratch despite far fewer parameters. Pretraining improves macro-F1 by 3-4 points, while the gap between the best pretrained CNN and transformer is below 0.5 points. Architectural choices nevertheless matter: ViT-S/8 outperforms ViT-S/16 and matches the four-times-larger ViT-B/16 at a quarter of the parameters, showing that finer tokenisation benefits small cell crops. Conversely, layer-wise learning-rate decay, central to the EVA-02 fine-tuning recipe, degrades performance, highlighting that transfer heuristics from natural-image recognition may not generalise to microscopy. Finally, knowledge distillation substantially improves the deployment frontier: compact EfficientNet-B0 students distilled from teacher councils outperform every individually trained backbone, including the EfficientNet-B5 and EVA-02 teachers. Overall, our results show that rigorous control of pretraining and evaluation is essential for interpreting biomedical architecture benchmarks, while distillation may be a more effective route to practical single-cell classification than architecture choice alone.
Read the full abstract
Number Theory
Universal norms and de Rham representations
The purpose of this note is to show that the computation of the universal norms in Iwasawa cohomology for the $p$-adic cyclotomic extension implies an analogous result for any $p$-adic extension that contains the cyclotomic extension. In addition, we prove a twisted cyclotomic variant, which applies to all Lubin–Tate extensions.
Read the full abstract
Geometric Topology
Computation of unknotting numbers: which knot breaks the Bernhard-Jablan Conjecture
We determine $2,540$ unknotting numbers of prime knots with at most $13$ crossings, that are unknown in the KnotInfo snapshot of 9 September 2026. The lower-bound calculations use Heegaard Floer correction-term obstructions and Greene's spanning-tree model. We also give explicit crossing-change constructions for upper bounds. For $959$ alternating knots with range $[2,3]$, we verify that no crossing change in a minimal diagram gives a knot of unknotting number one. We determine the unknotting numbers of all four knots in Brittenham and Hermiller's construction and thereby identify 13n3370 as an explicit counterexample to the original Bernhard-Jablan conjecture.
Read the full abstract
Machine Learning
Exact Degeneracy Under Balanced k-Shot Sampling:Consequences for Small-Sample Discriminant Analysis on LLM Embeddings
Balanced k-shot sampling, drawing exactly k labeled examples per class, is the standard protocol in few-shot evaluation. We show that it induces an exact degeneracy in Kernelized Linear Principal Component Discriminant Analysis (KLPCDA), a family of seven supervised dimensionality reduction methods. The within-class operator becomes exactly a scaled orthogonal projector, so two variants have all signal eigenvalues equal and a third has a zero operator, and the truncated pseudo-inverse used to implement them removes exactly the subspace that contains all between-class information. We verify this to float64 precision on sentence embeddings and on residual-stream activations of a decoder-only model, and find that the accuracies of the truncated variants change with the number of BLAS threads, by up to 12 percentage points (57 for the variant with a zero operator). Replacing the truncation with a ridge selected within the support set, we evaluate few-shot text classification on frozen embeddings (d up to 4096) over four datasets, three embedding models, and three trained baselines. The regularized formulation outperforms the best original variant in 45 of 48 settings, but in general not a tuned linear probe or nearest centroid on bidirectional encoders. Its gains are largest on E5-Mistral, the only decoder-based embedding model, and come from regularized within-class whitening rather than from KLPCDA: shrinkage LDA outperforms both baselines in all 16 E5-Mistral settings, and pre-specified tests show that the effect depends on the embedding model rather than on its dimension. E5-Mistral's remaining gap is largely an estimation-efficiency effect: increasing k to 30-50 reduces the probe's gap to a bidirectional encoder by more than 80%. We give five guidelines limited to the conditions tested and release the code and per-seed results.
Read the full abstract
General Economics
How an Economy Shrinks in Space: Concavity-on-Jobs and Upward Consolidation under Demographic Decline
When a country's population declines, the aggregate economy appears to contract on the intensive margin: industrial diversity intact, every industry a little smaller. At the regional level, contraction is uneven and takes the extensive form: entire industries disappear, one after another. The relevant unit is the city: industries are nested by size - the hierarchy property of industrial location - each viable only above a minimum population. Necessity industries' thresholds bunch at the low end, so a city's industry count - and its jobs - is sharply concave in size (concavity on jobs). A modest loss pushes a small city below many thresholds at once; a large core sheds a few specialized industries, one at a time. Lost industries consolidate upward to the next city large enough to host them; for the worker it means a step down to a lower-paid local job. To recover that income, workers move up to the apex - the only city hosting the full industry range. Studying Japan - two decades ahead of the OECD, Tokyo at its apex - with worker-level panel data on the young workers who carry the migration, a wage regression in real, housing-inclusive wages identifies a Tokyo-bound migration incentive that varies by origin, following concavity on jobs.
Read the full abstract
Dynamical Systems
A general endomorphism of $\mathbb{P}^k$ has trivial iterated centralizer
Fix integers $k,d\geq2$. Let $\mathrm{End}_d^k$ denote the parameter space of endomorphisms of $\mathbb{P}^k$ of algebraic degree $d$. We prove that there exists a dense Zariski open subset $U_{d,k}\subset\mathrm{End}_d^k$, defined over $\mathbb{Q}$, such that, for every $f\in U_{d,k}(\mathbb{C})$, every dominant rational self-map of $\mathbb{P}^k$ commuting with an iterate of $f$ is itself an iterate of $f$. The key intermediate result classifies dominant rational semiconjugacies to iterates of $f$ from normal projective varieties of dimension $k$. The proof uses finite-level monodromy and its action on the rooted preimage tree of a point outside the branch locus of an iterate of $f$. An analogous centralizer theorem holds generically for regular polynomial endomorphisms of $\mathbb{C}^k$. These results extend several theorems of Pakovich to higher dimensions. We give two applications. We first classify periodic dominant correspondences between a general endomorphism and an arbitrary endomorphism of degree at least two. We then show that, for a general $f\in\mathrm{End}_d^k$, an irreducible hypersurface of $\mathbb{P}^k$ is $f$-special in the sense of Ghioca–Tucker and DeMarco–Mavraki if and only if it is $f$-preperiodic.
Read the full abstract
Optimization and Control
MPFA: A Pareto Front Approximation Method for Riemannian Bi-objective Optimization
We propose a Pareto front approximation (MPFA) method for smooth bi-objective optimization problems on Riemannian manifolds based on a Hermite interpolation technique. Compared with the existing multiobjective optimization numerical algorithms, the proposed method can generate a continuous approximate Pareto front without multiple initial points. We establish convergence of the proposed method and analyze the approximation error of the resulting Pareto front. Numerical experiments on several test problems demonstrate that the proposed approach can effectively approximate the Pareto front with high accuracy and reasonable computational cost. Furthermore, the method is applied to a bi-objective formulation of sparse principal component analysis, illustrating its practical applicability in data analysis problems.
Read the full abstract
Computers and Society
With a Thermomix You Lose the Ability to Cook: A Kitchen Machine Analogy for Applications of Generative AI in Education
The rapid adoption of generative AI tools such as ChatGPT has sparked intense debate about their risks and opportunities for education, as well as the ways researchers should investigate them. In this paper, we approach these discussions through an analogy with the Thermomix, a smart kitchen appliance that has similarly provoked both enthusiasm and critique. By mapping Thermomix use cases onto examples of learning with generative AI, and situating them within the ICAP and SAMR frameworks, we show how different modes of tool use can either support or undermine meaningful engagement and learning. The Thermomix metaphor underscores that the central question is not whether learners employ AI, but how such use shapes their learning processes. In doing so, we provide a conceptual lens for researchers and practitioners to critically examine - and more effectively guide - the integration of generative AI into educational practice.
Read the full abstract
Machine Learning (Statistics)
A Unifying Perspective on Probabilities as Model Predictions
Although probabilistic statements are ubiquitous, foundational disagreements persist about their understanding, as exemplified by debates between Bayesians and frequentists; moreover, it is unclear when and why acting on them actually leads to desirable outcomes. Here, we argue that every probability is the output of a prediction method, that is, it depends on both a particular way of constructing abstractions and a way of transforming them into predictions. Through this, we provide a unifying perspective on supposedly different kinds of probabilities and show that even supposedly objective ones are model-dependent. We demonstrate that when a finite calibration criterion is met, one can anticipate the distribution of utilities for a given policy and inform successful decision-making on finite sets of events. Based on the notion of prediction methods, inductive arguments, and the probability calculus, we explain the feasibility of the calibration criterion in many settings. Overall, we develop a coherent perspective on probabilities and their use, connecting key intuitions behind other interpretations along the way.