Search arXiv⌕ Search

SEARCH · Search arXiv

Search Search arXiv

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

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,315 records · Page 73Linked to original sources
arXiv2026

Error-correction Decoding for 2D Convolutional Codes with a Generalized Viterbi Algorithm

Two-dimensional (2D) convolutional codes are a generalization of one-dimensional (1D) convolutional codes and are naturally suitable for encoding two-dimensional data. Although their theoretical properties are well established, practical applications are limited due to the lack of error-correction decoding algorithms. While the Viterbi algorithm provides an efficient decoding method for 1D convolutional codes, no analogous approach exists for the 2D case. In this paper, we introduce a decoding algorithm for 2D convolutional codes, called the 2D Viterbi algorithm, extending the principles of the classical Viterbi algorithm to the two-dimensional setting. The presented 2D Viterbi algorithm works with linear systems representations of 2D convolutional codes and its complexity depends on the dimension of the corresponding 2D linear system. Hence, we also provide several algorithms to obtain 2D realizations of small dimension.

cs.IT↗
arXiv2026

The Fer expansion of a driven spin under $\mathfrak{su}(2)$ operations cannot terminate beyond first order

The Fer expansion writes the propagator of a driven quantum system as an ordered product of exponentials, each generated by the time integral of an effective Hamiltonian that governs the evolution left over by the previous factors. It can be stopped after finitely many factors whenever this leftover evolution is known in closed form, for instance when the effective Hamiltonian commutes with itself at all times, or precesses about a fixed axis with constant angular velocity. For $\mathfrak{su}(2)$, in any representation, from a qubit to a spin of any size, we resum the Fer recursion into a closed map on curves in $\mathbb{R}^3$. We prove that, for a real-analytic drive, in both cases the expansion stops beyond the first factor only trivially, when the first exponential is already exact.

quant-ph↗
arXiv2026

A Discrete Amplitude-to-Time Conversion Circuit for TDC-Based Readout of Fast Scintillation Detectors

Fast timing measurements with scintillation detectors commonly employ leading-edge discriminators and time-to-digital converters (TDCs). Many applications also require an energy-related observable for estimating the deposited energy and correcting time-walk effects. Time-over-threshold (ToT) is frequently used for this purpose, but its relationship to pulse amplitude can be severely degraded by undershoot, ringing, and other distortions of the detector pulse tail. This work presents a discrete analog front-end circuit for amplitude-to-time conversion that is insensitive to pulse-tail distortions. An analog multiplexer and a high-speed comparator implement a sample-hold-discharge stage: upon leading-edge detection, the detector pulse amplitude is sampled near its peak and stored on a capacitor, which is subsequently discharged at a controlled rate. The resulting digital pulse width encodes the sampled amplitude, while the leading edge preserves the timing information. The circuit was designed for a short recovery time and a total dead time below 100~ns, making it suitable for high-rate detector readout. The approach enables simultaneous timing and amplitude measurements through a single TDC channel. An eight-channel front-end card implementing this concept was characterized using generated test pulses. The measured transfer function is approximately linear up to input amplitudes of 0.5~V and extends beyond 2~V, with a low-amplitude conversion gain of approximately 27~ps/mV. The estimated relative amplitude precision falls below \SI{0.5}{\percent} for input amplitudes above 0.5~V. These results demonstrate the feasibility of a low-complexity, multichannel amplitude-to-time converter for TDC-based detector readout, offering an alternative to conventional ToT measurement when pulse-tail distortions compromise amplitude estimation.

physics.ins-det↗
arXiv2026

Adaptive Stopping of Streamed Reasoning for Language Model Agents

Reasoning-enabled language model agents can incur substantial computation before acting, although the reasoning needed varies across decisions. Existing stopping signals, such as confidence and intermediate-answer stability, can vary in effectiveness across reasoning models; signals obtained from hidden states or token probabilities also require internal access that may be unavailable or impractical. We introduce StreamStop, which learns model-specific stopping policies offline from prefix-level action agreement. A frozen prefix reader encodes available reasoning text for a lightweight stop head, allowing stopping without access to the reasoning model's internal signals. We construct and publicly release a dataset of agent reasoning trajectories with prefix-level action annotations. These annotations compare each prefix action with the full-reasoning reference action, enabling offline comparison of stopping rules and exact expected-return optimization of a discounted action-agreement objective. Conditioning the head on the discount factor allows the agreement-token trade-off to be adjusted at inference without retraining. In closed-loop evaluations on WebShop and ScienceWorld, StreamStop achieves higher observed success than most validation-selected fixed-budget settings while using fewer reasoning tokens than Full thinking. On WebShop with Qwen3.8-27B, it uses 42% fewer reasoning tokens per episode with similar observed task success.

cs.CL↗
arXiv2026

Beyond the Frame: Redundancy, Not Trajectory, Carries the Privacy Leak in Video Unlearning

Class unlearning for video segmentation aims to ensure that a removed category cannot be reconstructed from a deployed model's outputs. Current leakage evaluation algorithms inspect one frame at a time, and we show that this understates leakage by a wide margin: an attacker that averages a model's output field over a two-second clip recovers the forgotten class from every method we evaluate on VSPW and most on Virtual KITTI~2, including models whose per-frame predictions are visually indistinguishable from a retrain oracle (a model trained from scratch without the forgotten class). We attribute the recoverable signal to cross-frame redundancy rather than object trajectory, establish this with a time-shuffled control that leaves leakage intact, and introduce PACE, an unlearning method that suppresses redundancy directly using an exactly permutation-invariant pooled adversarial discriminator and a two-sided temporal consistency objective. Across VSPW and Virtual KITTI~2, PACE lowers pooled attacker AP to $18.79$ (\texttt{person}) and $12.10$ (\texttt{car}) on VSPW and $9.11$ (\texttt{car}) on Virtual KITTI~2, achieving on the latter the strongest suppression among verdict-safe methods retaining usable utility, while retaining $19.8$ mIoU on VSPW \texttt{car} against at most $8.3$ for any baseline achieving comparable erasure.

cs.CV↗
arXiv2026

Annealed correlation and quenched noise sensitivity in biased deep neural networks

In earlier work by Jonasson, Steif and Zetterqvist (2023), fully connected deep neural networks with layers of width $n$, i.i.d.\ standard Gaussian weights, sign activations and no bias terms were shown to be annealed and quenched noise sensitive with respect to any joint probability distribution $\mathbb{Q}^{(n)}$ on the input $\boldsymbolω$ and its perturbation $\boldsymbolη$ satisfying a mild positive-correlation condition and $\mathbb{Q}^{(n)}(\boldsymbolη \in \{\boldsymbolω,-\boldsymbolω\}) \to 0$. Here we add independent Gaussian biases of the natural order and show that quenched noise sensitivity is retained over a broad range of depths, indeed under the slightly weaker assumption that $\mathbb{Q}^{(n)}(\boldsymbolω = \boldsymbolη) \to 0$. Thus, subject to the positive-correlation condition, quenched noise sensitivity holds under the minimal requirement that the perturbed input is asymptotically unlikely to coincide with the original input. At the same time, the annealed covariance remains bounded away from zero and is entirely accounted for by fluctuations in the network-dependent output means.

stat.ML↗
arXiv2026

Radial uniqueness and nondegeneracy for the Robin Lane-Emden equation

In this paper, we study positive radial solutions of $-Δu=u^p$ in a ball $B_R\subset\mathbb{R}^N$, subject to the Robin boundary condition $\partial_νu+βu=0$, where $N\ge2$ and $β>0$. For every Sobolev subcritical exponent $p>1$, we prove the existence of a unique positive radial solution and its nondegeneracy with respect to radial perturbations. The proof relies on the strict monotonicity of the logarithmic slope of the normalized Lane-Emden profile, which follows from the radial Pohozaev identity. When $N\ge3$, the same argument also recovers the classical critical existence range $0<βR<N-2$ and proves radial nondegeneracy throughout this interval. At $βR=(N-2)/2$, we exhibit $N$ linearly independent nonradial solutions of the linearized problem.

math.AP↗
arXiv2026

Understanding the formation of alpha-Fe bimodality in the Milky Way using constrained cosmological simulations of the Local Group

We use a novel suite of constrained cosmological simulations of the Local Group to investigate the formation of alpha--Fe bimodalities in stellar discs of Milky Way analogues. In agreement with past studies using the Auriga model, we show that in our simulations the high-alpha sequence forms in a merger-induced starburst, which further quenches star formation, allowing the new stellar populations to form at low [alpha/Fe]. Producing a bimodality that is prominent at large radii, with a significant [alpha/Fe] gap and overlapping metallicity as in the Milky Way, requires the quenching of star formation across all radii where the bimodality is observed. The quenching can be achieved either with a galaxy-wide starburst, or with a retrograde merger reducing angular momentum in the outer disc and allowing more efficient dilution by the metal-poor gas in the circumgalactic medium (CGM). We show that mergers can also tilt the disc, changing the direction of galactic outflows and enriching a larger volume of the CGM with galactic metals, leading to low [alpha/Fe] scatter in the subsequently formed low-alpha sequence. We argue that the slopes in [alpha/Fe] of the two alpha sequences are crucial in narrowing down the mechanism that leads to their formation. We additionally uncover an indirect role that mergers play in enhancing the accretion of metal-poor gas onto galaxies. The smoothly accreted gas can dominate the gas content in the CGM and aid in the formation of distinct chemical sequences, especially at the low-metallicity end.

astro-ph.GA↗
arXiv2026

Recast the Past: Stable Long Video Generation with Full Context Memory

Long-horizon video generation requires retaining an ever-expanding history, but conditioning on its growing token sequence is costly. Existing methods restrict or compress historical context, potentially losing information needed for future generation. This work proposes ReCam, which groups local spatiotemporal blocks of historical latents into vectors and projects them directly into the high-dimensional hidden space of a Diffusion Transformer. The model learns to use the resulting memory tokens through native self-attention, reducing token count while retaining block-level access to the full history. To mitigate temporal drift during autoregressive rollout, we further introduce dual-pointer denoising with separate noise schedules for recent context and target frames. Among the evaluated methods, ReCam achieves the highest VBench-Long overall score of 80.1 and competitive static-scene revisitation performance on R2M-Bench. It also leads on all four consistency metrics for off-screen subject reappearance, including clothing consistency of 4.09 versus 3.03 for the best baseline.

cs.CV↗
arXiv2026

Displaced e+e- Signatures from an X17-Mediated Inelastic Dark-Sector Cascade

We investigate displaced electron-positron signatures from an inelastic dark sector mediated by a light vector boson associated with the X17 anomaly. A heavy neutral fermion decays into a lighter stable state and the mediator, whose subsequent decay produces a single visible vertex after two stages of neutral propagation. We calculate the decay width, displacement distribution, lepton kinematics, and production rate in electro-positron collisions, including vector polarization. At Belle II, stronger dark-sector couplings increase production but require finely adjusted masses to preserve a macroscopic decay length. We quantify off-shell contributions and assess the validity of the two-body approximation. We show that the electron-positron pair cannot satisfy simultaneously the moment and opening-angle cuts of an existing search proposal for the configuration studied. Less restrictive selections allow nonzero kinematic acceptance. Our results clarify the experimental requirements for searching these signatures at Belle II.

hep-ph↗
arXiv2026

Elements of the Green ring related to Alperin's weight conjecture

J. L. Alperin introduced his celebrated weight conjecture, a major question in the modular representation theory of finite groups, which has now been studied for several decades. Alperin viewed this conjecture as the beginning of a ``Lie theory for the general finite group", as the title of the article indicates. This paper is intended as a tribute to Alperin's insight. In this note, we place an emphasis on projective modules and putting the numerical questions arising from the weight conjecture into a ``module form" within the appropriate Green ring of RG, when G is a finite group, and R is a suitable complete discrete valuation ring with residue field of prime characteristic p. We believe that this work moves in the direction of a structural approach to Alperin's weight conjecture.

math.RT↗
arXiv2026

On Johnson-Lindenstrauss Transforms for Byzantine-Robust Optimization

The growing scale of data and models in machine learning calls for efficient optimization methods. Distributed optimization addresses this challenge, with compression mechanisms helping to reduce communication costs. A classical tool for dimensionality reduction is the Johnson-Lindenstrauss (JL) transform, which is simple to implement and offers strong geometric guarantees. In particular, JL transforms approximately preserve inner products and pairwise distances between vectors with high probability. In this paper, we exploit this property in Byzantine-robust optimization, where worker messages are weighted according to their alignment with a trusted reference gradient. We further support our theoretical findings with experimental results.

cs.LG↗
arXiv2026

When Momentum Helps Polyak--Ruppert Averaging

Polyak--Ruppert averaging gives SGD and momentum methods the same asymptotic covariance. We characterize how fast finite runs reach it and how far they can be parallelized. For strongly convex quadratics with additive noise, we prove an online--offline bridge: the empirical risk minimizer on the samples of a tail average is one Newton step from the tail-averaged query point of any online procedure. For SGD, heavy ball, and Nesterov's method with constant momentum, this step is a boundary term at the two ends of the tail and splits the gap to the offline estimator into center, covariance, and shape errors. Necessary and sufficient conditions show that, on the least curved direction, the center error sets the pace: SGD needs $m\ggκ^2$ rounds but momentum only $m\ggκ$, where $κ$ is the condition number, so at sample size $N$ the critical batch size grows from $Θ(\sqrt{N}/κ)$ to $Θ(\sqrt{N/κ})$. The covariance error separates the momentum methods: heavy ball has relative variance excess $Θ(κ^{3/2}/m)$ along the stiffest direction, whereas Nesterov has relative error $O(\sqrtκ/m)$ in every direction. The shape error is method-independent by a Berry--Esseen bound uniform in $κ$ and batch size, so confidence intervals are valid when the other two errors vanish. We further examine how the conclusions change under heavy-tailed or state-dependent noise and clipping, and establish a first-order comparison with ERM for nonlinear Nesterov averaging under explicit conditions.

cs.LG↗
arXiv2026

Non-minimal unimodular inflation revisited: Swampland diagnostics and CMB constraints

We study non-minimal unimodular inflation in the Einstein frame; the effective potential takes the form $U(φ)=U_0(1+ξφ^2/\mpl^2)^{-2}$. We consider the canonical field map and the inflationary observables directly from the Einstein-frame action. The exact potential slow-roll parameters, the leading scalar running, the scalar-amplitude normalization, the e-fold relation, and the end-of-inflation condition are obtained analytically. The model has a nontrivial structural tension: an $ε_V=1$ exit requires $ξ>1/2$, whereas the near-origin slow-roll regime requires $ξ\ll1$ and gives $n_s\simeq1-8ξ$. We next test three distinct swampland diagnostics. The further-refined de Sitter conjecture (FRDSSC) introduces the independent conjectural parameters $a$, $b$, and $q$, while the scalar weak gravity conjecture (SWGC) and its strong form (SSWGC) contain no analogous tunable conjecture parameter in the single-field realization adopted here. All three criteria are evaluated using the canonically normalized Einstein-frame scalar. Analytic limits show that the SWGC is violated at the stationary point for every nonzero $ξ$, whereas the SSWGC at that point requires $ξ\gtrsim0.0643$. In the large-field limit both scalar weak-gravity diagnostics reduce to the same threshold $ξ\ge0.1$. Numerical scans then show that the observationally relevant ACT/P-ACT region does not satisfy either scalar weak-gravity criterion, while the FRDSSC can be satisfied for $a$ close to unity. Consequently, the minimal model can reproduce acceptable $(n_s,r)$ at isolated field values, but this does not constitute a complete 50-60 e-fold inflationary history or simultaneous satisfaction of all three conjectures.

hep-th↗
arXiv2026

Mean-field limit and Vlasov equation for a Gaussian-thermostatted Lorentz gas

We study N particles moving in a dispersing billiard under an electric field and coupled through a Gaussian thermostat. The corresponding mean-field equation is a Vlasov-type equation with specular reflection. We reformulate the equation as a fixed point problem for the time-integrated thermostat coefficient. This yields existence of weak solutions for every absolutely continuous initial datum of finite energy and, for bounded initial densities supported on velocities that are large compared with the field, uniqueness of the weak solution; under the finite horizon assumption, the upper bound on the velocities can be replaced by an exponential tail of the initial density. In these regimes, we prove convergence of the empirical measures of the particle system to the solution of the kinetic equation. In general, convergence holds along subsequences.

math.AP↗
arXiv2026

Building Telecom Language Models with RSOD: Domain Data, Specialist Post-Training, Memory-Conscious Distillation, and Networking Evaluation

Telecom language models must combine broad networking knowledge with specialist behavior across routing, broadband services, access technologies, and operational diagnostics while remaining practical to deploy as a single model. We present an end-to-end framework that builds source-grounded synthetic training data from structured networking taxonomies and expert graphs, materializes 212,559 routed training records across fourteen networking routes, trains six frozen specialist LoRA teachers over one shared continued-pretrained backbone, and consolidates their behavior into one deployable student adapter using Routed Specialist On-Policy Distillation (RSOD). In RSOD, the committed student generates each trajectory, the route-assigned specialist teacher-forces the exact native token sequence, compact teacher hidden states are reconstructed through the shared frozen output head, and only the student LoRA is updated. A controlled reconstruction validation shows mean absolute logit error of 7.26 x 10^-6 with 100% top-1 agreement between native and reconstructed teacher logits. On four RTX 6000 Ada GPUs, the completed 2,000-step run produced 16,000 trajectories with zero observed student-version or token- ID mismatches. Under the primary route-conditioned evaluation over 2,850 held-out questions, the raw Meta-Llama-3.1-8B-Instruct model scores 12.8661/25 micro (13.0172/25 macro), the combined routed six-teacher system scores 21.7641/25 micro, and the consolidated RSOD student, released as Llama-CalixCLM v1, reaches 21.7112/25 micro (21.8006/25 macro) on the same five dimension LLM-judge rubric. Under a separate neutral deployment-like prompt, the same student reaches 20.0028/25 micro (20.0778/25 macro). The consolidated student remains within 0.0529 micro-score points of the combined routed-teacher system while retaining a numerically validated and auditable train-serve execution path.

cs.AI↗
arXiv2026

Phase shifts and fluctuations from motional, relativistic, and dark-matter effects in ion-trap clocks

Ions possess both center-of-mass motion and internal states, rendering trapped-ion optical clocks susceptible to perturbations of both degrees of freedom, which results in systematic shifts and fluctuations with motional and relativistic origins. We develop a quantum theory that connects these effects to filter functions determined by the employed pulse sequences and show how the interrogation can be tailored to suppress or enhance specific perturbations. Our treatment includes laser frequency noise and motional effects such as recoil and Doppler shifts, micromotion and excess micromotion, as well as state-dependent trap imperfections and accelerations, e.g., caused by Stark shifts. We further account for relativistic effects including time dilation and the mass defect, and show how the same framework can be applied to hypothetical dark-matter signals. By considering ensemble-averaged phases and their variances, we characterize both systematic shifts and fluctuations, while asynchronous operation relative to the Paul-trap drive allows motional fluctuations to average down between runs. Our results provide a unified framework for characterizing and controlling the systematic and statistical contributions that determine the trueness and precision of trapped-ion optical clocks.

quant-ph↗
arXiv2026

Learning Tactile-Aware Hand Representations from Human Hand-Object Interaction Data

Tactile representation learning for dexterous manipulation often requires substantial robot-side data collection, limiting its scalability across different hands and sensors. We explore whether large-scale human hand-object interaction (HOI) data can provide a useful prior for robotic tactile learning. To this end, we propose HuTact, a framework that pretrains tactile-aware hand representations from human HOI data without requiring real tactile signals during pretraining. HuTact transfers the interaction structure learned from human hand motions to robotic tactile observations and can be further adapted with a small amount of unlabeled robot interaction data. We evaluate HuTact on tactile perception and manipulation tasks, including grasp prediction, slip detection, object classification, and tactile-conditioned policy learning. Across these tasks, human HOI pretraining improves data efficiency and provides a strong initialization for downstream learning. Notably, with HOI pretraining followed by midtraining on unlabeled robot play data, HuTact achieves 88.26% Macro-F1 on grasp prediction using only 25\% of the labeled downstream data, comparable to training from scratch on the full labeled dataset (88.23%). These results suggest that human interaction data can serve as a scalable complementary source for robotic tactile representation learning.

cs.RO↗