arXiv:2606.08671v3 Announce Type: replace
Abstract: Agent skills extend language-model agents with task-specific procedures, scripts, and references, but the tasks and environments they target continually change. Existing methods improve skills in bounded runs and retain only the final artifact, discarding the decision history that later agents need to interpret prior revisions, evaluations, and rejected alternatives. We introduce SkillHone, a harness for continual agent skill evolution grounded in persistent decision history. SkillHone pairs skill revisions with evaluation-side evidence that supplies practice feedback, recording structured histories of diagnoses, revisions, evidence, and outcomes. Role-separated subagents run candidate skills on practice probes with redacted reporting and propose revisions informed by prior decisions, enabling cross-session refinement without rediscovering past rationale. On deep-research benchmarks, SkillHone runs without a pre-integrated search stack and outperforms the commercially backed deep-research agent by 15.8 points on GAIA and 3.2 points on WebWalkerQA-EN, while also exceeding prior skill-evolution methods. We further deploy SkillHone on internal tool-mediated analysis scenarios, where it improves accuracy by an average of 18.8 points across seven settings.
Science Journals
arXiv:2606.11032v2 Announce Type: replace
Abstract: Existing deep learning models for Positron Emission Tomography (PET) image denoising often suffer from severe performance degradation under distribution shifts, fundamentally restricting their robust clinical deployment. This lack of generalization stems from the conventional paradigm of fixed-parameter models that cannot adapt to variations in test data (e.g., dose levels or scanner types) after training. To overcome this limitation and achieve robust generalization, we introduce U-TTT, a novel U-shaped model that integrates Test-Time Training (TTT) layers to dynamically adjust model parameters during inference through self-supervision, thereby adapting to the specific characteristics of each test instance. Furthermore, to comprehensively capture the complex degradations of 3D PET data, U-TTT features a dual-domain adaptation mechanism comprising a Spatial Test-Time Training (S-TTT) layer and a Frequency Test-Time Training (F-TTT) layer. The S-TTT layer captures and corrects spatial structural degradations, while the F-TTT layer suppresses global noise spectra and restores delicate high-frequency details. Extensive experiments demonstrate that U-TTT achieves state-of-the-art PET denoising performance and exhibits superior generalization under challenging distribution shifts, including both unseen dose levels and unseen scanners. Our code will be available at https://github.com/Yaziwel/U-TTT.
arXiv:2606.12569v3 Announce Type: replace
Abstract: We present eCREAM-MedCorpus, a new and unique large-scale dataset of clinical notes produced in Emergency Departments of Italian hospitals. The corpus, in its current version, is composed of approximately 4 million clinical notes fully anonymized, covering diverse phases of patient care during the stay in the emergency department. In addition, a subset of about six thousand notes has been manually annotated by clinical experts through a structured Case Report Form (CRF) containing 132 items relevant for two patient situations in emergency departments, dyspnea and loss of consciousness. Items may assume numerical values (e.g., for blood saturation), categorical (e.g., for level of consciousness ), binary (e.g., for presence of traumas), and mixed value types. The annotation process involved multiple clinicians and underwent iterative revision to resolve ambiguities in item formulation, resulting in a richly structured (although high imbalanced) resource. The dataset aims to fill a relevant gap of data able to support both the development and the use of Large Language Models in concrete medical applications. We describe the data collection protocol, the on-site anonymisation pipeline, corpus statistics, and the annotation scheme. Finally, we propose CRF-filling as a novel structured information extraction benchmark, and provide zero-shot baseline resulting from Gemma-27B and MedGemma-27B. To the best of our knowledge, eCREAM-MedCorpus is the largest freely available dataset of clinical notes existing for the Italian language.
arXiv:2606.12742v4 Announce Type: replace
Abstract: Wearable healthcare devices are the fastest-growing Internet of Things (IoT) sector. Many automated healthcare services rely on two crucial biological signals, namely ECG and EEG, which reflect the activity of the heart and brain, respectively. Although deep neural networks are considered the primary way to process and analyze these signals, the very tight energy and computational power constraints in wearable devices are far below the computational, energy, and memory bandwidth demands of DNN models, thereby impeding the deployment of deep learning in many practical wearable services. This paper investigates the feasibility of deploying state-of-the-art DNN models in resource-constrained wearable devices. Notably, we explore the trade-off between accuracy and computational complexity of DNNs when parameter quantization and electrode reduction methods are used. Our investigation centers on several state-of-the-art DNN models designed for EEG signal analysis, specifically for detecting epileptic seizures. Our findings demonstrate that, when applied judiciously, these techniques can significantly reduce the complexity of the DNNs under consideration with minimal adverse effects on accuracy. These results reveal the explicit trade-offs between accuracy and complexity reduction encountered when adapting DNN-based online EEG analysis for wearable devices.
arXiv:2502.12307v2 Announce Type: replace
Abstract: For a fixed alphabet $A$, an infinite sequence $X$ is said to be normal if every word $w$ over $A$ appears in $X$ with the same frequency as any other word of the same length. A classical result of Agafonov (1966) relates normality to finite automata as follows: a sequence $X$ is normal if and only if any subsequence of $X$ selected by a finite automaton is itself normal. Another theorem of Schnorr and Stimm (1972) gives an alternative characterization: a sequence $X$ is normal if and only if no gambler can win large amounts of money by betting on the sequence $X$ using a strategy that can be described by a finite automaton. Both of these theorems are established in the setting of deterministic finite automata. This raises the question as to whether they can be extended to the setting of probabilistic finite automata. In the case of the Agafonov theorem, this question was positively answered by L\'echine et al. (2024) in a restricted case of probabilistic automata with rational transition probabilities.
In this paper, we settle the full conjecture by proving that both the Agafonov and the Schnorr-Stimm theorems hold true for arbitrary probabilistic automata. Specifically, we show that a sequence $X$ is normal if and only if any probabilistic automaton selects a normal subsequence of $X$ with probability $1$. We also show that a sequence $X$ is normal if and only if a probabilistic finite-state gambler fails to win on $X$ with probability $1$.
arXiv:2602.21710v2 Announce Type: replace
Abstract: An extended formulation of out-of-time-ordered correlators (OTOCs), which quantify noncommutative operator growth and information scrambling in quantum many-body systems, is developed for turbulence dynamics as a representative of non-canonical Hamiltonian systems. Based on the Wigner-Weyl transform and the Moyal bracket formalism, the semiclassical limit of OTOC for turbulent plasmas governed by the Hasegawa-Mima equation is derived as an ensemble-averaged squared Lie-Poisson bracket between two chosen functionals of the turbulent fields. The classical-limit OTOC provides a quantitative measure of how a variational perturbation applied to one functional propagates across scales in the turbulent dynamics and how it affects another functional at a later time, thereby capturing scale-dependent or field-dependent transfer processes. In a quasilinear approximation with a strong zonal flow, we provide a closed analytic expression of the classical-limit OTOC to characterize the interaction between zonal and non-zonal modes. An asymptotic analysis shows that the OTOC grows quadratically at early time, while in the long-time strong-shear regime it approaches a finite saturated value with an inverse-square algebraic dependence. This behavior is attributed to zonal-flow shearing, which rapidly scrambles the non-zonal perturbation toward higher wavenumbers, thereby reducing the low-wavenumber non-zonal content that can feed back onto large-scale zonal modes.
arXiv:2607.05599v1 Announce Type: new
Abstract: This paper develops a computationally efficient framework for reachability analysis of transmission-level power system dynamics with synchronous generators, grid-forming and grid-following inverters, and uncertain power injections/withdrawals. Starting from reduced-order device models and a frequency-divider representation, we derive a linear ordinary-differential-equation model suitable for efficient reachable-set computation under bounded disturbances across network buses. The proposed reachability method combines interval reachability and contraction-based bounds to construct certified over-approximations for the linear ordinary-differential-equation model. A real Jordan transformation separates non-oscillatory modes, handled through a linear embedding system, from oscillatory modes, enclosed using contraction-based ball bounds. Numerical experiments on a modified IEEE 39-bus system validate the reachable tubes against high-fidelity electromagnetic-transient (EMT) simulations, and demonstrate multi-second reachable sets computed in sub-second time.
arXiv:2607.05440v1 Announce Type: cross
Abstract: Recommending the correct set of energy conservation measures (ECMs) for a building is a structured, multi-label prediction problem in which a task-specific supervised model has weak training signal and a general language model has no grounding in the local building stock. We study this problem on 10,422 real New York City Local Law 87 (LL87) energy-audit records, taking as ground truth the set of ECM categories that certified auditors actually recommended. We make four contributions. First, we establish that energy-use-intensity (EUI) prediction - the upstream task - is effectively solved by tree ensembles: across fifteen trained models, a stacking ensemble reaches a coefficient of determination R^2 = 0.757, and every one of six neural architectures is outperformed by gradient-boosted trees. Second, we show that the framing of the recommendation task dominates model choice: recasting ECM recommendation as 19-way multi-label classification rather than single-label categorization lifts a gradient-boosted-tree baseline from a previously reported 25.9% accuracy to a micro-F1 of 0.571. Third, we benchmark eight large language models (LLMs) from four providers in a 2x2 design that independently toggles retrieval grounding and explicit reasoning, scoring each arm on per-label F1, U.S.-dollar cost per building, and latency; retrieval-augmented generation (RAG) improves micro-F1 by +0.11 to +0.20 on every model, while explicit reasoning yields no measurable accuracy change (-0.018 to +0.010) at up to 8.4x the cost. Fourth, we show LLMs systematically over-recommend - high recall, low precision - and that retrieval closes the gap chiefly by improving precision. A 70-billion-parameter open-weight model with a fifteen-line nearest-neighbor retrieval step reaches 0.511 micro-F1 at $0.00032 per building, comparable to a frontier model at roughly 10.1x lower cost.
arXiv:2606.13757v2 Announce Type: replace
Abstract: Large language models (LLMs) are increasingly deployed in automated code-review systems, where their approvals can determine which code is merged into shared repositories. However, it is unclear whether review agents can detect vulnerability-introducing code when an attacker controls both the code change and the persuasive Pull Request (PR) narrative designed to mask it. We introduce SEVRA-BENCH (Social Engineering of Vulnerabilities in Review Agents), a benchmark that measures how often a review agent approves such adversarial PR s. Each PR in SEVRA-BENCH is built from a historical commit that fixed a vulnerability. We automatically reverse that fix to extract the original vulnerable code, and submit the resulting code change as a PR wrapped in one of 15 social-engineering framings. To test review-agent resilience to narrative manipulation, these framings vary dimensions such as supporting evidence, conveyed urgency, signals of prior approval, and appeals to authority. SEVRA-BENCH evaluates a retained challenge split of roughly 1000 adversarial PRs drawn from publicly disclosed vulnerability fixes across the top 10 entries of the MITRE's 2025 most dangerous software weaknesses. Evaluating 8 review agents against this benchmark, we reveal that review agents are susceptible to narrative manipulation, exposing a significant gap in security capabilities.
arXiv:2606.14048v3 Announce Type: replace
Abstract: World action models (WAMs) have recently shown promise in jointly modeling future observations and executable robot actions. However, most existing WAMs still operate in 2D video or latent spaces, where visually plausible rollouts miss the 3D spatial constraints and occluded contact geometry required for precise manipulation. While geometric foundation models offer strong priors for recovering dense 3D structure and motion from visual observations, forcing WAMs to predict the dense 4D representation introduces costly geometric decoding and slows down causal action generation. To address the trade-off, we present WAM4D, a fast 4D world action model that uses lightweight spatial register tokens as training-time future-depth readouts to transfer pretrained geometric priors into a causal video-action transformer, then removes the register branch for lightweight action inference. To prevent non-causal shortcuts, we further design causal mixture attention for the Mixture-of-Transformers (MoT) WAM backbone, defining modality-specific visibility among video, action, and geometry tokens. Comprehensive experiments on RoboTwin 2.0 and challenging real-world manipulation tasks show that WAM4D improves spatial consistency and achieves competitive action prediction while maintaining efficient inference.
arXiv:2606.18118v2 Announce Type: replace
Abstract: The Chebfun software system is a Matlab extension essentially based on representations of (piece-wise) smooth one-variable functions by expansions in Chebyshev polynomials. One of Chebfun's attractive features is the ability to provide solutions to nonlinear boundary value problems (BVP) with accuracy close to the machine precision. This is done by the chebop class which provides automatic solutions by performing linearizations with a Newton method in function spaces of the nonlinear BVP, automatic differentiation, and using Fast Fourier Transform computations for the coefficients of the Chebyshev polynomials. A drawback of chebop automatic approach is the possible lack of convergence of the Newton method if the initial guess is not close enough to the exact solution. An explicit functional linearization done for each particular shape of the differential operator (i.e. without automatic differentiation) proves to be more robust than the chebop class and allows an enlargement of the range of convergence. Another alternative is the implicit L-scheme (quasi-Newton approach with derivatives replaced by suitable positive constants L), with a much simpler implementation and globally convergent. While chebop is the easiest way to solve the BVP, provided that it converges, the last two approaches largely overcome the convergence issues, yielding accurate solutions to a wide class of steady-state one-dimensional problems governed by Richards' equation. Chebfun2 and Chebfun3, which at the current stage cannot solve BVPs, provide efficient tools for accuracy and convergence assessments of the non-steady solutions in one or two spatial dimensions obtained by classical discretization schemes.
arXiv:2606.19990v2 Announce Type: replace
Abstract: While RL has become a promising tool for refining world models, existing methods largely rely on conservative rollouts near the training distribution, limiting exploration, behavioral diversity, and richer dynamic discovery. In this work, we challenge this conservative paradigm. We argue that the core limitation is not exploration itself, but the lack of reliable verification strategies to support broader exploration. Without reliable verification, expanded exploration becomes highly susceptible to reward hacking, where policies exploit imperfect rewards without achieving genuine improvement. To evaluate this motivation, we instantiate our method in embodied world models, where physical plausibility, and task completion provide a rigorous testbed for scalable RL under complex dynamics. On the verification side, we introduce Reward as an Agent, an agentic reward framework that actively evaluates generated behaviors to provide robust reward signals and mitigate reward hacking under distribution shifts. On the exploration side, we introduce Dynamic-Aware Rollout Diversification through DynDiff-GRPO, which explicitly expands action-space exploration to diversify trajectories, broaden state-action coverage, and encourage richer embodied behaviors beyond conservative rollout regimes. By unifying Reward as an Agent with DynDiff-GRPO, we enable RL on a more reliable reward foundation with substantially diversified sampling, effectively mitigating reward hacking while yielding significant accuracy gains across multiple open-source world models, thereby demonstrating that broader exploration can scale successfully when grounded in robust verification.
arXiv:2606.20031v3 Announce Type: replace
Abstract: Dynamic environmental changes, confined workspaces, and stringent real-time constraints make pathfinding in Robotic Mobile Fulfillment Systems (RMFS) a challenging problem for conventional search- and rule-based methods, which typically suffer from high computational complexity and long decision latency. While reinforcement learning (RL) has emerged as a powerful alternative, deploying learned policies with extreme energy efficiency on resource-constrained hardware remains an open challenge. We present SDQN-RMFS, an end-to-end framework that achieves high-fidelity deployment of an RL-trained policy from a full-precision artificial neural network (ANN) through to a neuromorphic chip. By computing only when triggered by sparse events, this framework unlocks ultra-low-power RMFS pathfinding. Our full-stack pipeline operates as follows: an ANN policy is first efficiently trained via a collision-allowing strategy to densify informative trajectories, and then converted into a spiking neural network (SNN) via a hard-label knowledge distillation approach. This effectively addresses the output distribution mismatch, preserving policy capability across the ANN-to-SNN pipeline while substantially reducing inference latency. Hardware experiments demonstrate up to 11,281$\times$ energy savings and a nearly two-fold reduction in latency compared to a high-performance GPU baseline, while maintaining decision quality on par with the original trained policy. These results establish physical neuromorphic inference as a practical and energy-sustainable pathway for large-scale RMFS operations.
arXiv:2606.20416v2 Announce Type: replace
Abstract: Diffusion models rely heavily on explicit timestep embeddings to modulate the denoising process across various noise scales. In this work, we challenge the necessity of these temporal signals by analyzing their impact on U-Net and Diffusion Transformer architectures. Beyond empirical evidence, we provide a theoretical framework demonstrating that, under certain conditions, the global minimizer of the diffusion training objective can be achieved without explicit timestep conditioning. Our findings reveal a surprising robustness when timestep embeddings are completely removed. Extensive ablation studies on the CelebA and CIFAR-10 datasets show that these time-agnostic models can maintain high structural fidelity and even surpass their conditioned counterparts in competitive metrics, including FID, precision, and recall. Our analysis suggests these architectures can implicitly infer noise scales from the corrupted input under specific assumptions, rendering explicit temporal conditioning redundant. This study challenges long-standing temporal conditioning paradigms and paves the way for more efficient and structurally focused generative architectures.
arXiv:2605.13196v2 Announce Type: replace
Abstract: Direct numerical simulation of microscale fluid--structure interactions in multicomponent and multiphase flows requires methods that can represent moving boundaries together with fields constrained to evolving interfaces. Diffuse-domain methods (DDMs) address this geometric difficulty by replacing sharp surfaces with diffuse volumetric representations on regular computational domains. Here we formulate DDMs using Onsager's variational principle. Instead of extending sharp-interface equations and boundary conditions term by term, we embed sharp-surface free-energy and dissipation functionals into the bulk through a diffuse surface delta density and derive the governing equations from the Rayleighian. The framework distinguishes balance-law fields, internal nonconserved order parameters, and kinematic or constitutive rate variables. It also clarifies a key moving-surface distinction: conserved surface densities are transported by the full material surface velocity, whereas explicitly tangential vector and tensor internal variables require projected objective or co-rotational rates within their admissible tangential state spaces. For scalar transport on rigid and deformable interfaces, and for interfacial hydrodynamics near rigid walls, the formulation recovers established DDM models and their sharp-interface limits. The same variational construction yields coupled diffuse-domain models for multicomponent deformable vesicles with surface viscosity, tangential slip, and finite areal compressibility, and for active shells carrying chemical and tangential vector order. These results provide a unified route to thermodynamically consistent passive DDMs for interfacial and surface dynamics, while allowing active stresses through active work power. The framework is relevant to soft matter, microfluidic interfaces, biological membranes, and morphogenetic surface dynamics.
arXiv:2605.16147v2 Announce Type: replace
Abstract: Vision Transformers (ViTs) are known to exhibit high-norm patch-token outliers that degrade feature map quality, a problem effectively mitigated by register tokens. As diffusion models increasingly adopt transformer architectures and move toward pixel-space training, they become closer in form to ViTs, raising the question of whether register tokens are also useful for Diffusion Transformers (DiTs). In this work, we show that DiTs differ from ViTs in a key respect: they do not exhibit patch-token outliers but still benefit from registers. Interestingly, registers are more effective in pixel-space DiTs than in latent-space DiTs. By analyzing intermediate representations, we find that register tokens produce cleaner feature maps at high noise levels, which may contribute to their effectiveness in pixel-space generation. We further observe that recent pixel-space DiT architectures implicitly incorporate register-like mechanisms, which may partially account for their strong empirical performance. Motivated by these observations, we propose Register Guidance, a technique that amplifies the contribution of register tokens responsible for improving visual structure and coherence.
arXiv:2510.07514v2 Announce Type: replace
Abstract: Inverse Kinematics (IK) is a core problem in robotics, in which joint configurations are found to achieve a (collision-free) desired end-effector pose. Modern IK solvers face a fundamental trade-off: analytical methods are fast but lack generality, while numerical optimization-based methods are broadly applicable but prone to local minima and high computational costs. To overcome this challenge, we introduce HJCD-IK, a GPU-accelerated, sampling-based hybrid solver. By pairing a novel orientation-aware greedy coordinate descent initialization with Jacobian-based polishing and a parallel collision filter, our method achieves up to order-of-magnitude gains in speed and accuracy over state-of-the-art solvers, consistently finding collision-free solutions on the accuracy-latency Pareto frontier, while producing a diverse distribution of high-quality samples. We validate our solver on a physical Franka manipulator and release our code open-source.
arXiv:2311.07461v4 Announce Type: replace
Abstract: Autonomous systems (AS) often rely on Deep Neural Network (DNN) classifiers to operate in complex and dynamically changing environments. However, during operation, these classifiers may encounter domains that differ from those seen during development, causing performance degradation under distribution shift. Removing systems from operation for labelled data collection and retraining is often impractical, particularly when adaptation must occur quickly and at scale. This paper introduces DIRA-SS, a self-supervised extension of Dynamic Incremental Regularised Adaptation (DIRA) that enables online domain adaptation using only a small number of unlabelled target-domain samples. DIRA-SS augments an existing classifier with an auxiliary retraining branch and adapts the shared feature representation through a rotation-prediction task, while elastic weight consolidation regularises important source-domain parameters to reduce destructive updates. This allows the model to benefit from transfer learning without requiring classification labels during operation. We evaluate DIRA-SS on CIFAR-10C, CIFAR-100C, and ImageNet-C using ResNet architectures under severe common corruptions. The results show that DIRA-SS substantially improves performance over the non-adapted source model, achieves accuracy close to the supervised DIRA method, and outperforms existing unsupervised test-time adaptation baselines on ImageNet-C when using only 100 target-domain samples.
arXiv:2607.05423v1 Announce Type: new
Abstract: Upper Confidence Bound (UCB) algorithms guarantee sublinear regret for agents learning unknown stochastic environments, yet the same principle that makes them statistically efficient (optimism in the face of uncertainty) induces a predictable strategic vulnerability against an omniscient adaptive leader. Classical strong Stackelberg equilibrium (SSE) assumes that the follower immediately best-responds to the leader's committed mixed action; it therefore supplies no mechanism-design prescription for a leader facing a boundedly rational follower who constructs and acts on empirical reward histories. We formalize this conflict in a finite-horizon repeated Stackelberg game and give exact constructive proofs for a deceptive leader mechanism. In a honeypot phase, the leader pays a finite signaling cost to inflate the UCB index of a designated follower action. In a trap phase, the leader switches to a selfish action distribution while the follower remains locked into the designated action because the manipulated empirical history and exploration bonus dominate competing indices. Under explicit separation and payoff assumptions, the leader's cumulative utility strictly exceeds the classical SSE ceiling, and the manipulation cost is bounded by a regret calculation of order $O(\sqrt{T\ln T})$. The results identify a formal incompatibility between static equilibrium prescriptions and dynamically learned empirical incentives.
arXiv:2607.05456v1 Announce Type: new
Abstract: While recent advances in large language models have enabled end-to-end automated manuscript generation, existing systems suffer from three critical deficiencies: (i) generated claims are not deterministically grounded in verifiable literature, (ii) experimental results are frequently fabricated rather than executed, and (iii) there exists no standardized, multi-dimensional framework to assess whether AI-generated manuscripts meet the quality and rigor required for real-world publication. We present Prompt-to-Paper, a multi-agent framework that directly addresses this evaluation gap through three integrated innovations. First, a deterministic retrieval-augmented generation pipeline with section-aware relevance scoring and snowball citation expansion grounds every claim in a verifiable corpus of 60--100 papers. Second, an autonomous coding agent executes real computational biology experiments replacing synthetic outputs with genuine numerical results. Third, an eight-dimensional automated quality scorer, benchmarked with approximate reference statistics from published papers and augmented with explicit hallucination penalties, provides standardized, reproducible quality assessments. The quality-driven improvement loop uses a context-rich reviser that routes each iteration to one of three researcher actions and fires a deep research cycle every ten iterations to re-run experiments and re-manuscript from stronger outputs. We validate the system on five bioinformatics case studies; all five cases compiled submission-formatted PDFs with zero out-of-range citations. The improvement loop raises manuscript quality by an average of +17.96 points on a 0--100 scale (maximum +26.04. As partial external checks, a human reviewer scored the five manuscripts at an average of 7.0 out of 10. Complete manuscripts are produced at approximately 0.31 USD per paper.
arXiv:2507.01577v3 Announce Type: replace
Abstract: We consider interpolation from the viewpoint of fully automated theorem proving in first-order logic as a general core technique for mechanized knowledge processing. For Craig interpolation, our focus is on the two-stage approach, where first an essentially propositional ground interpolant is calculated that is then lifted to a quantified first-order formula. We discuss two possibilities to obtain a ground interpolant from a proof: with clausal tableaux, and with resolution. Established preprocessing techniques for first-order proving can also be applied for Craig interpolation if they are restricted in specific ways. Equality encodings from automated reasoning justify strengthened variations of Craig interpolation. Contributions to Craig interpolation that emerged from automated reasoning include variations for logics used in databases and logic programming. As an approach to uniform interpolation we introduce second-order quantifier elimination with examples and describe the basic algorithms DLS and SCAN.
arXiv:2606.22511v2 Announce Type: replace
Abstract: In open-ended generation, LLMs frequently fall into the "likelihood trap", marked by repetitive degeneration and vocabulary dullness, creating a discrepancy between machine-generated and human-written text. While post-hoc tail truncation (e.g., Top-$p$, Min-$p$) avoids sampling from the unreliable tail, it can over-sample from the uncalibrated head and misalign generation with human lexical preferences; fixed scalar repetition penalties likewise ignore variation in logit scale across inference steps, potentially disrupting semantic coherence. To address both limitations, we propose Variance-Calibrated Modulation (VCM), a training-free pre-decoding intervention that reshapes the probability distribution before truncation through two dynamic mechanisms: (1) Contextual Searchlight via PMI, which suppresses global stopwords while elevating context-evoked tokens, and (2) Adaptive Self-Debiasing, which uses real-time logit standard deviation for scale-invariant penalization. Across open-ended generation, factual QA, and mathematical reasoning, VCM consistently mitigates the likelihood trap. With negligible computational overhead, VCM integrates with existing decoding strategies, improving diversity, coherence, and, particularly at higher decoding temperatures, reasoning accuracy.
arXiv:2606.22807v2 Announce Type: replace
Abstract: As retrieval systems scale, high-quality reranking becomes increasingly important. However, most existing rerankers, whether encoder-based or decoder-based, jointly encode the query and passage, tightly coupling their computation and limiting deployment efficiency as well as flexibility. We present KaLM-Reranker-V1, a fast but not late-interaction (FBNL) reranker that decouples query and passage computation while retaining expressive relevance modeling. Built on an encoder-decoder architecture, KaLM-Reranker-V1 uses the encoder to pre-encode passages with Matryoshka embedding pooling, while the decoder models the system instruction, user instruction, and query intent; cross-attention then captures relevance between the query context and passage representations. This design makes KaLM-Reranker-V1 efficient through decoupled passage encoding, yet not late interaction, by preserving rich relevance modeling through cross-attention. We instantiate KaLM-Reranker-V1 in three sizes, Nano, Small, and Large, with 0.27B, 1B, and 4B activated parameters, respectively. Extensive experiments on BEIR, MIRACL, and LMEB demonstrate that KaLM-Reranker-V1 achieves strong reranking performance with superior efficiency. On BEIR, KaLM-Reranker-V1 achieves state-of-the-art performance, on par with strong industrial models such as the Qwen3-Reranker series; on MIRACL, despite not being extensively trained on multilingual data, KaLM-Reranker-V1 still shows excellent reranking performance. Moreover, on LMEB, reranking models demonstrate a clear advantage, with even the 0.27B Nano model remaining competitive with 7-12B embedding models.
arXiv:2606.24234v2 Announce Type: replace
Abstract: Autonomous robotic inspection in maritime environments presents unique challenges for Visual Place Recognition (VPR) due to cross-scene perceptual shifts. Robots navigating ship-borne environments must transition between visually distinct domains: open decks with sparse textures and severe illumination changes, and enclosed cabins with repetitive structures and high visual ambiguity. Existing VPR methods, designed primarily for urban or indoor scenes, fail to generalize reliably across these starkly different scenarios. To address this, we propose ProteusVPR, a two-stage retrieval-refinement framework. The first stage employs any standard VPR model for initial image retrieval. The second stage introduces a geometric-visual estimation network that fuses the retrieved image with two temporally preceding frames, incorporating geometric descriptors, a local affine coordinate system, and camera azimuth encoding to achieve precise localization. To support this task, we introduce the XHZ dataset, an 8K-panoramic ship-borne dataset collected from an operational vessel, featuring multi-floor cabin structures, deck transition zones, and strict query-database separation for rigorous evaluation. Extensive experiments on the XHZ dataset demonstrate that ProteusVPR consistently improves the localization accuracy across multiple VPR backbones, reducing mean localization error by over 60\% on average and that ProteusVPR offers an effective and robust solution for precise visual localization in challenging, cross-scene maritime environments.
arXiv:2606.27106v2 Announce Type: replace
Abstract: We present a novel approach for applying Large Language Models (LLMs) to threat assessment in the context of foreign peacekeeping missions. Building on the PINPOINT project and its use case, the EU Monitoring Mission in Georgia, we combine an interdisciplinary risk-model with OSINT-based media collection and LLM-supported threat extraction. The proposed workflow maps media contents to mission-relevant threats, extracts structured information and applies several additional LLM-based processing steps to improve relevance and grounding. An evaluation of threats extracted from media documents shows high agreement between automatically generated results and human judgment for core aspects such as threat and mission relevance. These results indicate that LLMs provide a promising approach to support analysts in the context of peacekeeping missions.