Search arXiv⌕ Search

arXiv subjects

Search papers

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

At least 1,477 records · Page 82Linked to original sources
arXiv2026

When Do Self-Supervised Representations Help Speech Restoration?

Self-supervised learning (SSL) representations are widely used in speech restoration (SR), and the prevailing trend is SSL feature cleansing: fine-tuning the SSL to recover clean features from degraded speech, then conditioning the restoration model on them. However, the representational capability of SSL when encoding degraded speech has been underexplored, and we ask whether cleansing always benefits SR training. Across SSL encoders and perceptual quality proxies, we discover perceptual correlation collapse---under severe degradation, cleansing degraded SSL features toward clean ones misaligns with the SR objective. We hypothesize the restoration backbone can supply the SSL with clues to resolve this collapse, since its intermediate layers already produce partially restored features. We therefore propose Early Restoration Injection (ERI), which routes intermediate features from the restoration backbone into the SSL through a lightweight conditioning network. ERI excels over the unconditioned baseline and, on the VoiceFixer benchmark, achieves competitive performance among multiple baselines.

cs.SD↗
arXiv2026

Designing sustainable high-entropy alloys directly from mixed oxides: A thermodynamic perspective

Hydrogen-based direct reduction (HyDR) of mixed oxide precursors offers an alternative route to produce sustainable alloys, bypassing the sequential extraction, refining, and remelting steps that dominate the carbon footprint of conventional metallurgy and individual metal production. However, the thermodynamic limits and most effective H$_2$/H$_2$O ratio needed for full metallization of multicomponent oxides are not yet established. This is because the formation of metallic and oxide solid solutions alters the chemical activity of both the reduced metallic phases and the unreduced oxides, coupling their equilibria in ways that cannot be deduced from the stability of the pure individual oxides. Herein, we develop a CALPHAD-based thermodynamic framework to establish the compoistion-temperature-H$_2$/H$_2$O window for complete metallization of CoCrFeMnNi-based oxide mixtures. Mn and Cr generally increase oxide stability, while Co, Fe, and Ni decrease it, allowing reduction at lower H$_2$/H$_2$O ratios. The effect of each element is not fixed; it depends on the starting mixed oxide composition and can reverse when the last oxide to reduce changes. Furthermore, we show that the reduction pathway complexity does not correlate with the difficulty of removing the final oxygen: a system with few intermediate transformation steps can still terminate as a highly stable residual oxide, whereas extensive partial reduction may leave the stability of the final residual oxide essentially unchanged. By linking residual-oxide stability to the H$_2$/H$_2$O ratio, the framework identifies the thermodynamic limits and provides guidelines for designing sustainable multicomponent alloys directly from mixed oxide precursors.

cond-mat.mtrl-sci↗
arXiv2026

Uncertainty estimation of incoming comet trajectories with a thermophysical non-gravitational force model for Comet Interceptor encounter planning

The European Space Agency Comet Interceptor mission is designed to fly-by an as-yet undetermined incoming long-period or dynamically-new comet, which will require an accurate prediction of its trajectory. Cometary orbits are perturbed from purely gravitational trajectories by non-gravitational accelerations (NGAs) stemming from their outgassing, however, and these must be taken into account in the navigation. We here test the ability of ground-based astrometry and standard NGA-fitting techniques to predict the position and velocity of incoming comets perturbed by a Monte Carlo sampled thermophysical acceleration model. We find that, due to the difficulties of predicting NGA evolution as well as the limitations of reasonably simulated astrometry, predicted accuracies are unlikely to meet the mission goals of 1000 km and 1 m s$^{-1}$ in position and velocity, sixty days before intercept. We instead recommend expanded uncertainties of $10,000$ km and several m s$^{-1}$. Our NGA fits struggle to reproduce the reference accelerations and trajectories when outgassing is rising steeply at a few AU-incoming, but with more observations up until perihelion, do eventually converge, leading to good estimates of many of the underlying acceleration parameters even for half an orbital arc.

astro-ph.EP↗
arXiv2026

Lip-Focused Audio-Visual Consistency Detects Speech Content Manipulation beyond Coarse Synchronization

Speech-content manipulation changes what a visible person appears to say by replacing the speech track and, in stronger attacks, regenerating mouth motion to match the forged utterance. Such attacks are difficult to expose with texture artifacts or coarse lip-sync scores because modern retalking models explicitly optimize audio-visual synchrony. We test the narrower hypothesis that speech replacement still perturbs the local relationship between acoustic dynamics and mouth motion after speaker identity or coarse synchrony is controlled. To isolate this signal, we introduce SyncSpeech, an LRS3-derived benchmark containing real videos, audio-splicing cheapfakes and Wav2Lip retalking attacks under same-speaker and different-speaker replacement. We then introduce C-MIND, a lip-focused detector that aligns audio and mouth-motion representations at frame level and models their agreement and disagreement without transcripts, automatic speech recognition features or an unmanipulated reference. On SyncSpeech, pretrained SyncLip-C achieves 95.813% scenario-averaged accuracy and 0.9943 AUC, outperforming sentence-level fusion, full-face inputs and a reproduced SyncNet baseline under the reported protocol. When trained on the corresponding FakeAVCeleb training split, C-MIND correctly classifies 278 of 280 samples in the selected unseen-speaker test subset, whereas accuracy falls to 82.775% on unseen Video-Retalking attacks. These results identify lip-focused speech-articulation consistency as a useful but generator-sensitive forensic signal within the evaluated settings.

eess.AS↗
arXiv2026

MOER: Mix-Policy Optimization under Expert Prefix Guidance with Selective Rollback

Mix-policy optimization incorporates off-policy expert data into the reinforcement learning (RL) loop to transcend the capability boundary of on-policy RL. However, existing methods either provide static full-trace guidance that suppresses self-exploration, or adopt expert prefix guidance without correctness guarantees, leading to suboptimal performance or even training collapse. In this paper, we propose MOER, a mix-policy optimization framework that harmonizes expert guidance with self-exploration. MOER injects a decaying expert chain-of-thought prefix into the prompt and lets the model complete the remaining reasoning, with the prefix length progressively annealed to encourage self-exploration. A selective rollback mechanism replaces failed completions with the original full expert trace, guaranteeing the correctness of every off-policy trajectory. Furthermore, a harmonized mix-policy objective optimizes on-policy tokens with a GRPO loss while supervising both expert prefix and completion tokens with a trust-region SFT loss, substantially stabilizing training. Extensive experiments show that MOER achieves the highest average scores in mathematical and general domain reasoning tasks, outperforming all basic and mix-policy baselines, and generalizes consistently across diverse tasks, model scales, and base model categories. Codes are available at https://github.com/CityU-MLO/MOER.

cs.LG↗
arXiv2026

Hear Every Sound Object: Rescuing Overlooked Sources in Auditory Scene Analysis

Auditory scene analysis (ASA) models extract information about individual sound sources from a complex auditory mixture. Recent object-based ASA models process the mixture through a sequence of modules and split it into source-wise features (i.e., object features) only at the last layer. However, substantial disparities in performance across sources limit overall performance. Our analysis using centered kernel alignment (CKA) reveals that CKA scores correlate strongly with per-source performance, with their disparities widening from intermediate to deeper layers. These findings motivate the early split and exchange of object features to address source-wise performance disparities before they widen in deeper layers. To this end, we propose PARASITE, a plug-in refinement architecture that splits object features at earlier layers and progressively refines them through sequential layers. Each layer integrates object features separated at different depths using Object-wise Gated Fusion (OGF) and then refines them using Windowed Cross-Object Attention (WCOA). We introduce the DeepASA++ model, which combines PARASITE with an ASA backbone and delivers state-of-the-art results on both ASA2 and STARSS23. DeepASA++ achieves a SELD score of 0.240 on STARSS23 and surpasses prior multi-stage inference architectures, including chain-of-inference.

eess.AS↗
arXiv2026

HimmBO: High-Dimensional Interactive Model-Merging Bayesian Optimization for Image Generation

Fine-tuning has become a common way to customize text-to-image models, leading to community collections of over 100K lightweight adapters. Adapters sharing a base model can be merged with continuous coefficients. This turns a large collection into a high-dimensional continuous design space, which recent work explores interactively using preferential Bayesian optimization (PBO). However, existing approaches are limited to a few dozen preselected adapters and require expensive image generation in every interaction round. We introduce HimmBO, a two-stage method that scales preference-driven adapter merging to much larger collections while reducing image-generation cost. Its screening stage recovers the active adapter set through level-set estimation over pre-generated images, requiring no image generation. The refinement stage then optimizes merge coefficients in the recovered low-dimensional subspace. Both stages use a style-aware kernel, allowing preference information to transfer between visually similar adapters. In simulated-user image-matching experiments, HimmBO achieves higher final similarity than all comparisons while reaching a given similarity with under one tenth of the image generations required by prior work. Its performance remains stable as the collection grows to 1080 adapters, while competing methods degrade substantially.

cs.CV↗
arXiv2026

ATLAS: A Multi-Agent Framework for Lifting Binary to Canonical Compiler IR

Lifting stripped, heavily optimized binaries to a canonical, pre-optimization LLVM intermediate representation (IR) is essential for downstream security analysis and legacy software migration when source code is unavailable. However, aggressive compiler optimizations rewrite control and data flow, remove structural metadata, and introduce type-erased low-level idioms, causing current lifters to produce malformed or semantically divergent IR. We present ATLAS, a multi-agent framework designed to reconstruct formally valid and canonical LLVM IR from optimized disassembly.ATLAS combines (i) Structure-aware, region-level retrieval, which utilizes pseudo-probe attribution to align optimized instructions with canonical IR exemplars, and (ii) a collaborative refinement loop that closes the gap between probabilistic generation and deterministic compiler constraints. Concretely, a Lift agent generates an initial IR draft conditioned on retrieved exemplars, an Advisor agent interprets LLVM verifier/compilation diagnostics and static checks to propose localized repair actions, and a Validation agent applies the edits, re-runs toolchain verification and (when available) runtime checks, and feeds the resulting diagnostics back to the Advisor. The agents iterate until the IR is verifier-clean and behaviorally consistent, or a budget limit is reached. On HumanEval-Decompile and MBPP, ATLAS outperforms seven baselines on lifting LLVM IR, achieving 91.8% re-executability and 89.4% CFG similarity across ARM, x86-64, and RISC-V binaries. Moreover, our evaluation demonstrates that ATLAS also outperforms the Binary to C to IR pipelines in pointer-flow precision and downstream analyzability. By recovering a well-formed, machine-centric representation, ATLAS enables practical cross-ISA recompilation and verifiable binary-only migration workflows.

cs.SE↗
arXiv2026

Dual medial axes in finite-dimensional normed spaces

We study medial axes of nonempty closed sets in finite-dimensional real normed spaces whose norms are $C^1$ away from the origin. Using the differential of the norm, we define the dual $k$-medial axis by requiring the contact covectors associated with nearest points to have affine dimension at least $k-1$. In dimension $n$, we prove that this set is $(n-k+1)$-rectifiable for every $1\le k\le n+1$, and hence has Hausdorff dimension at most $n-k+1$. We give a direct geometric proof and characterize the dual medial axes through the superdifferential of the squared distance function, obtaining an alternative proof from a theorem of Alberti, Ambrosio, and Cannarsa on singularities of semiconvex functions. Under the additional assumption of strict convexity, the ordinary and dual medial axes coincide for $k=2,3$, yielding the corresponding rectifiability results for ordinary medial axes. For $k=4$, however, the ordinary rectifiability conclusion can fail even for norms arbitrarily close to Euclidean: we construct norms on $\mathbb R^3$ with smooth positively curved unit spheres arbitrarily $C^\infty$-close to the Euclidean sphere, together with four-point sets whose ordinary $4$-medial axes contain nondegenerate line segments.

math.MG↗
arXiv2026

Relativistic Dynamical Evolution of Kilonova Spectral Emission

We develop a relativistic dynamical model for the thermal evolution of expanding kilonova ejecta, in which the effective temperature is determined by the ejecta mass, opacity, velocity, and radioactive heating rate. The resulting thermal evolution follows a broken power law as the ejecta transition from the optically thick to the optically thin regime. We incorporate this thermal history into a relativistic spectral calculation that accounts for the equal-arrival-time surface geometry and Doppler boosting, enabling us to model the observed kilonova spectrum and constrain the underlying ejecta properties. We investigate the effects of the model parameters and relativistic corrections on the observed temperature and spectral shape, and find that light-travel-time delays and Doppler shifts can produce significant spectral distortions at high ejecta velocities. We apply the model to the early-time X-shooter spectrum of the kilonova AT2017gfo and infer the ejecta mass, opacity, and velocity through Markov chain Monte Carlo (MCMC) fitting with a Student's $t$ likelihood. We find that relativistic effects become increasingly important for fast ejecta ($v \gtrsim 0.5c$), particularly at large viewing angles, where the observed temperature can differ substantially from the rest-frame effective temperature. These results highlight the importance of incorporating relativistic and equal-arrival-time effects when interpreting early-time kilonova spectra, especially for rapidly expanding ejecta.

astro-ph.HE↗
arXiv2026

One Teacher, Many Views: Reducing Reference Sensitivity in On-Policy Self-Distillation

On-policy self-distillation (OPSD) improves reasoning by training a model on its own trajectories with dense token-level supervision from a privileged copy conditioned on a reference solution. Existing methods typically use a single teacher distribution conditioned on the complete reference. We identify a limitation of this formulation: a reference is only one realization of the reasoning process. Although it provides evidence about the answer, it also encodes an ordering, phrasing, and solution path. Full-reference conditioning can therefore produce correct targets that depend excessively on information unavailable to the student at inference. We show that replacing one valid reference with another, while holding the problem and student state fixed, can substantially alter the teacher distribution, a phenomenon we call reference-induced target instability. To address this instability, we introduce Monte Carlo Privileged Distillation (MCPD), which models privileged context as a distribution over partial reference views and distills the marginal teacher distribution. MCPD evaluates the teacher under multiple views and averages the resulting next-token distributions. We show that this probability-space average is the exact target induced by minimizing the expected teacher-to-student KL divergence over privileged views. Across mathematical and scientific reasoning benchmarks, we show that MCPD produces more reference-consistent supervision and consistently improves over full-reference OPSD.

cs.LG↗
arXiv2026

On the maximal index of $C_4^-$-free unbalanced signed graphs of prescribed size

The extremal problem of determining the graphs with maximum spectral radius among all graphs with a prescribed number of edges was resolved by Rowlinson [Linear Algebra Appl. 110 (1988) 43--53]. A signed graph $Γ=(G,σ)$ is called $C_k^-$-free if it contains no negative $k$-cycle as a signed subgraph. The index of a signed graph is defined as the largest eigenvalue of its adjacency matrix. This paper addresses the maximization of the index of unbalanced signed graphs. For signed graphs of a prescribed order, Wang and Lin [Discrete Appl. Math. 372 (2025) 164--172] determined the maximum index of $C_4^{-}$-free unbalanced signed graphs. Subsequently, Wang, Peng and Hou [Discrete Math. 349 (2026) 114871] characterized the unbalanced signed graphs of prescribed size with maximum index. This paper treats the edge spectral Turán-type problem for $C_4^-$-free unbalanced signed graphs. The all-positive signed graph with underlying graph $G$ is denoted by $(G,+)$. Evidently, $A((G,+))=A(G)$. In particular, $(H_{d,t},+)$ is the all-positive signed graph whose underlying graph is $H_{d,t}$, where $H_{d,0}=K_d$ and for $t>0$, $H_{d,t}$ is obtained from $K_d$ by adding a single vertex adjacent to exactly $t$ vertices of $K_d$. Let $m-3=\binom{d}{2}+t$ with $0\le t\le d-1$. Up to switching isomorphism and the addition of isolated vertices, the unique extremal unbalanced signed graph is obtained by attaching a negative triangle to a specified vertex of the all-positive signed graph $(H_{d,t},+)$.

math.CO↗
arXiv2026

Inherited, Not Invented: Tracing the Provenance of Wait in Reasoning Models

Reinforcement learning (RL) is credited with teaching reasoning models to write {wait} before reconsidering a step, yet large RL accuracy gains do not require the word. This view can confuse what RL discovers with what a model inherits from its training data. We trace the word through open models' training histories, reading it with label-free indices of frequency, position, form and usage and an injected-error test, all computed from sampled outputs. We find that zero-RL almost never coins the marker: on bases that lack the word, RL raises MATH-500 accuracy substantially while {wait} stays near absent. It arrives with training data: mid-training in OLMo-3, and fine-tuning on reasoning traces installs it in most models whose base lacks it. Where the base already uses it, both released RL-Zero recipes (OLMo-3, MiMo) multiply it, whereas KL-free GRPO from the same base does not; OLMo-3's recipe wears down its form and makes it a more sensitive but more liberal alarm; RL after supervised fine-tuning barely changes it, and our GRPO recipe erases a marker seeded by fine-tuning. Finally, we show that marker frequency alone does not indicate RL-induced reasoning and caution against LLM-as-a-judge revision metrics, as automated judges can reverse human-labelled trends.

cs.LG↗
arXiv2026

Spectral and edge conditions for negative Hamiltonian cycles in unbalanced signed graphs

Spectral conditions for Hamiltonian cycles are a fundamental subject in extremal spectral graph theory. For simple unsigned graphs, Fiedler and Nikiforov [Linear Algebra Appl., 432 (2010), 2170-2173] pioneered this direction, and Li and Ning [Linear Multilinear Algebra, 64 (2016), 2252-2269] later extended these results by introducing minimum degree parameters. In this paper, we study the spectral conditions for the existence of a negative Hamiltonian cycle in unbalanced signed graphs of order n > 8. We first determine the maximum index among all n-vertex unbalanced signed graphs containing no negative Hamiltonian cycle, and characterize the corresponding extremal graphs up to switching isomorphism. Furthermore, we establish an upper bound on the number of edges in unbalanced signed graphs of order n >= 7 with frustration index one, and classify all extremal graphs achieving this bound.

math.CO↗
arXiv2026

Tracing Subliminal Learning Through Pretraining

Large language models trained on data generated by other models can sometimes inherit traits unrelated to that data, a phenomenon called subliminal learning. We study both hard label distillation, where students train on numbers generated by a biased teacher, and on-policy soft label distillation, where students generate the data and the teacher's logits provide the learning signal. Using Pythia, PolyPythia, and DataDecide OLMo models, we find that transfer persists in both settings across differences in weight initialization, pretraining data order, pretraining corpus, model size, and architecture. Across pretraining checkpoints, the ability to transmit a preference and the ability to acquire one develop separately: some checkpoints are more effective teachers but weaker students, and vice versa. Consistent with this distinction, cross-model transfer can be strongly asymmetric in magnitude when teacher and student roles are reversed. To investigate what carries the preference signal, we construct synthetic number sequences that match the teacher's generated number frequencies but have no dependence on the prompt or preceding numbers and find that these still induce substantial transfer in the hard-label setting. In top-$K$ soft-label distillation, removing animal tokens from the loss reduces transfer, although not completely in the same-model setting. In our experiments, soft label distillation produces more consistent transfer with less extensive training than hard label distillation. The growing use of on-policy distillation in model post-training makes this a particularly relevant setting for assessing unintended preference transfer.

cs.LG↗
arXiv2026

From Perpetrators' Digital Footprints to Behavioral Communities: An Ontology-Driven Framework for Bengali Cyberbullying Analysis--- FULL VERSION

Cyberbullying involves behavioral and contextual patterns beyond isolated abusive messages. However, existing approaches remain fragmented across content detection, semantic representation, and network analysis, with limited interpretable perpetrator representations, multi-scale structural modeling, and explanations of behavior-driven similarity and influence propagation. These limitations hinder the identification of persistent perpetrators, characterization of behavioral communities, and assessment of potentially influential users. This paper proposes Ontology-Driven Cyberbullying Behavior and Network Analysis (ODCBNA), a perpetrator-centric framework integrating requirements-driven ontology design, knowledge graph construction, behavioral modeling, community analysis, and influence modeling. Using the Bengali Contextual and Diverse Cyberbullying (BenCoDiC) corpus as a use case, ODCBNA identifies $1{,}063$ users with at least three bullying comments or replies as perpetrators and represents each through a $64$-dimensional behavioral and contextual feature vector. Cosine similarity and $k$-nearest-neighbor modeling produce a weighted network with $27{,}554$ edges. Louvain analysis identifies six communities with a modularity of approximately $0.495$ and $27$ sub-communities. Feature-contribution analysis identifies the Video, Impact, and User Profile groups as the strongest contributors to behavioral similarity and community separation, while contextual dimensions support interpretation. Community-constrained influence analysis shows that IM Greedy achieves small positive gains in all six communities, although different seed-selection methods produce comparable diffusion spreads despite limited seed overlap. These findings provide an interpretable basis for understanding perpetrator behavior, community organization, and potential influence patterns in Bengali online environments.

cs.CL↗
arXiv2026

The Reaction of Financial Markets to the Releases of DeepSeek R1 and Kimi K3: A Critical Analysis

This paper compares financial-market reactions to the releases of DeepSeek R1 in January 2025 and Kimi K3 in July 2026. Both AI models were presented as credible Chinese competitors near the capability frontier, but their releases conveyed different economic information. DeepSeek R1 combined open weights with unusually low reported training and inference costs, prompting investors to reassess expected demand for computing infrastructure and producing large stock-price decreases among semiconductor and data-centre-related firms. Kimi K3 instead entered at broadly frontier prices and generated a market reaction concentrated more heavily among rival Chinese AI laboratories. Drawing on descriptive event-window evidence, model capability and cost comparisons, subsequent industry developments, and the historical diffusion of open-source software, the paper argues that open-weight models redistribute rents across the AI value chain rather than eliminating them. Incumbents remain protected by complementary assets including distribution, product bundling, switching costs, proprietary data, serving capacity, and contractual trust. The analysis suggests that investors and policy makers should assess risks looking not only at model capability, but also at the ownership and durability of complementary assets across the AI value chain.

econ.GN↗
arXiv2026

A sharp gradient-corrected Lieb--Thirring inequality in one dimension with applications to density functional theory

We first prove a sharp Lieb-Thirring inequality with semi-classical constant and gradient correction in one dimension. In the periodic setting, the corresponding inequality also includes a parity dependent shell-correction term, which appears to be new. Moreover, we also characterize the optimizers of these inequalities. In particular, the optimizing density profiles form a two-dimensional manifold, which we call here the Poincaré density manifold. We then apply these results to study the Thomas--Fermi--von Weizsäcker (TFW) approximation of the kinetic energy in Density Functional Theory. More precisely, we first obtain an explicit and non-asymptotic bound on the global and local error made by the TFW approximation. These bounds are controlled by the distance of the normalized density profile from the Poincaré density manifold times the number of particles. In particular, these bounds show that the TFW approximation is exact on the Poincaré manifold. We then show that, in the periodic setting, the distance between the Kohn-Sham potential and the inverse TFW potential in the natural dual Sobolev space can also be controlled by explicit density dependent constants times the distance to the Poincaré manifold. As a corollary, we prove that for a fixed density profile, the Kohn-Sham potential converges towards the inverse Thomas--Fermi potential in the high-density limit $N \rightarrow \infty$ with asymptotic rate of order $N^{-2}$. This rate is shown to be optimal.

math-ph↗