arXiv:2607.13940v1 Announce Type: new
Abstract: Personal health management unfolds over repeated encounters, yet most health AI systems treat each request in isolation. We developed HealthClaw, an open-source agent architecture that updates support as a person's routines, preferences, measurements and risks change. It separates shared safety rules and medical knowledge from private longitudinal memory containing profile facts, reusable procedures and episodic traces. After each episode, induction determines what should update the profile, revise a procedure, remain episodic or be excluded. We evaluated HealthClaw with a synthetic year-long benchmark and nine 200-case biomedical tasks. Across 900 longitudinal support probes, answer accuracy increased from 0.2% with current-query prompting to 45.7% with HealthClaw, while prompt-side context exposure was 71.7% lower than with full-history prompting. In 100 privacy probes, HealthClaw produced higher privacy-aware answer quality and fewer unsafe disclosures than both baselines. Across the biomedical tasks, the mean absolute gain in the task-specific primary metric was 27.0 percentage points, and seven gains remained significant after false-discovery-rate correction. These offline benchmarks support governed, self-evolving memory for longitudinal personal health agents, although clinical effectiveness requires prospective evaluation. HealthClaw is publicly available at https://github.com/HC-Guo/HealthClaw.
Science Journals
arXiv:2607.13941v1 Announce Type: new
Abstract: Video aesthetic assessment (VAA) aims to predict how aesthetically pleasing a video is, yet remains far less explored than other visual assessment tasks. Its progress is hindered not only by the scarcity of large-scale benchmarks, but also by the intrinsic subjectivity of aesthetic judgment, which is shaped by human perception. In this paper, we revisit VAA from a psychological perspective and propose \textit{Peak-End-Net}, a lightweight and interpretable framework inspired by the \textit{peak-end rule}, which suggests that people tend to judge a temporal experience mainly according to its salient moments and the ending. Building on this intuition, we first transfer knowledge from image aesthetic assessment (IAA) to VAA by introducing a pretrained IAA head to produce frame-wise aesthetic priors, which serve as surrogate signals for identifying aesthetically salient moments and guiding \textit{peak-end rule}-based temporal aggregation. To further capture how a video evolves aesthetically over time, we design an aesthetic rhythm encoder that models temporal progression beyond isolated moments. Additionally, we refine the overall assessment through a dynamic gated fusion mechanism to improve robustness under distribution shift. Our method is built on a frozen vision transformer (ViT) and requires only a small number of trainable parameters, making it scalable and parameter-efficient. Extensive experiments on two existing VAA benchmarks, including in-domain evaluation on VADB and cross-domain testing on DIVIDE-3K, demonstrate that our approach achieves state-of-the-art performance, affirming the value of psychologically grounded modeling for VAA. Our code and models are available at https://github.com/AMAP-ML/Peak-End-Net.
arXiv:2604.21448v3 Announce Type: replace-cross
Abstract: This work presents a unified viscoelastic-viscoplastic continuum framework for modeling rate-dependent granular flows across regimes. The formulation incorporates two distinct rate-dependent mechanisms, namely micro-inertia and viscoelastic dissipation, within a single continuum description. A central contribution is an explicit link between the coefficient of restitution and a continuum viscosity, derived from an analysis of wave attenuation in granular assemblies, thereby establishing a direct connection between particle-scale collision physics and macroscopic damping. This relation is introduced while retaining inertia-dependent plastic flow governed by the classical $\mu(I)$ rheology. The constitutive model is constructed by meticulously partitioning elastic and plastic responses within the model and corresponding stress-update routine, such that viscous dissipation governs wave propagation and collisional processes without altering the plastic flow rule. The framework is implemented within the material point method to simulate transient processes involving large deformations, material separation, and subsequent reconsolidation. A range of numerical examples, including steady, transient, vibrational, and impact-driven flows, demonstrates that the model captures wave propagation, diffusion, and rate-dependent granular behavior within a unified continuum setting.
arXiv:2607.00507v2 Announce Type: replace-cross
Abstract: Quantum contextuality is a vital non-classical resource, yet illuminating the precise mechanisms through which it enables unconditional computational advantages remains a challenge. We translate graph-theoretic formulations of contextuality into an unconditional quantum memory advantage for formal language recognition. We define a promise problem on an exclusivity graph $G$ where any classical finite automaton respecting exclusivity requires $N \ge \chi(G)$ memory states, whereas a QFA requires a memory of dimension $d = \xi(G)$. The gap between these bounds isolates a structural, information-theoretic incompatibility between classical and quantum descriptions that we term \textit{representational contextuality}. For Boolean orthogonality graphs, this exacts an exponential classical memory penalty ($d=\mathcal{O}(n)$ vs $N=2^{\Omega(n)}$). Finally, we demonstrate a sharp algorithmic phase transition: allowing the classical machine a finite confusability of mutually exclusive events reduces this exponential classical memory cost to $\mathcal{O}(n)$.
arXiv:2607.06597v2 Announce Type: replace-cross
Abstract: Public chest-radiograph (CXR) datasets are typically released with small, fixed label schemas such as CheXpert-14. However, the underlying free-text reports describe far more findings -- and which findings matter depends on the task, site, and reader. We release a pipeline that converts free-text reports into multi-label matrices and then reconfigures the label schema through dictionary edits rather than new inference passes, i.e., without relabeling the corpus. After this one-time pass, reconfiguring MIMIC-CXR (223K reports) from cached annotations takes 196 seconds with no API cost, compared to \$6.6K for an equivalent relabeling pass with Claude Opus 4.7. Using a 58-label taxonomy, we show that 43\% of CXR studies contain at least one finding outside CheXpert-14. Image probes trained on these labels match CheXpert-14 probes on shared targets while also reaching 0.78 AUROC on expert-reviewed long-tail labels that CheXpert-14 cannot represent. These results suggest a different unit of work for radiology labeling: once reports are structured, the label schema becomes a configuration to edit, not a corpus to relabel.
arXiv:2607.13618v1 Announce Type: new
Abstract: LLM agents are increasingly evaluated on multi-week decision tasks in which the state that drives cost is never directly observed. On such tasks the final cost cannot say why an agent failed: it may have misread the world, or read it correctly and still failed to act (the knowing-doing gap). Existing evaluations cannot separate these two failures; their reference policies either read privileged information the agent never sees, or are missing altogether. We introduce STOCKTAKE, a 26-week supply-chain replenishment benchmark built as a factored partially observable Markov decision process with six hidden factor processes, designed so that a fair reference policy is computable: an exact Bayes filter per factor drives a rollout policy on the identical observation stream the agent receives. Scoring each run between a symptom-blind base-stock floor (0) and this oracle (1) yields a skill score, and grading each week's written rationale yields a stated-belief detection lag and a knowing-doing rate, so state estimation and control are measured separately. On fifty seeds with curated stress profiles, Claude Sonnet 5, GPT-5.4, DeepSeek-V4-Pro, and Grok 4.5 detect 84-88% of hidden failures, typically within a week of onset, yet span skill scores from 0.62 to -0.23: two of the four end below the symptom-blind floor while naming factors slightly faster than the two that beat it. The failure has two faces. Where stress persists, 34-43% of correctly diagnosed stress weeks still end in stockout for every model, a rate that partly reflects the severity of the weeks models notice. That rate also runs opposite to skill: the two models under the floor stock out least on diagnosed weeks, so under-response is only one face of the gap, and their traces point to the other, responses whose cost exceeds what they protect. STOCKTAKE measures both directions of that failure.
arXiv:2607.13954v1 Announce Type: new
Abstract: We report an optimized synthesis and crystal growth protocol as well as structural, optical, and terahertz (THz) generation characteristics of the organic nonlinear optical crystal ZPAN ((Z)-1-(((4-Phenylamino)phenylamino)methylene)naphthalen-2(1H)-one). Noncentrosymmetric packing of ZPAN is most reliably achieved by slow evaporation from an acetone-based solvent system, producing long rectangular prisms (7 cm \times 1 cm \times 0.2 cm) with (010) as the main face and [001] as both the polar axis and the direction of elongated crystal growth. [001] is thus the most effective pump polarization direction for generating THz light via optical rectification. The THz generation characteristics of the developed ZPAN crystals are determined at different near-infrared irradiation wavelengths, crystal thicknesses, and pump powers, revealing that ZPAN can generate a peak-to-peak electric field of near 1 MV/cm with a smooth spectrum from 0.5-3.4 THz. Calculated second order nonlinear optical coefficients indicate that both the (100) and (010) faces are theoretically capable of THz generation, though (010) is consistently the dominant growth face. The large size of this face (7 cm \times 1 cm) makes it possible to scale laser power with aperture size, for use in extremely high-power laser systems.
arXiv:2607.13965v1 Announce Type: new
Abstract: Open source software is vulnerable to supply-chain attacks through transitive dependencies, especially malicious code injected into NPM packages. Existing detectors often inadequately model obfuscated behavior, overlook JavaScript's object-centric features, poorly coordinate static and dynamic analysis, and lose semantic information during behavior abstraction. We propose ProfMalPlus, a malicious NPM package detector combining object-sensitive behavior graphs with coordinated LLM reasoning over annotated code slices. It identifies installation commands and entry files, then constructs graphs capturing sensitive APIs, third-party calls, and unresolved calls. From these graphs, ProfMalPlus extracts security-relevant slices and adds inline static analysis evidence. Local judge agents independently assess each slice. Self-consistency consolidates repeated judgements to reduce LLM variance, while a global judge synthesizes their reports into an entry-level verdict. For undetermined cases, a router selects either third-party enrichment, which adds registry derived module and method semantics, or dynamic augmentation, which executes the package in a sandbox to resolve runtime dependent behavior. The enriched evidence is fed back for reassessment. Finally, a localization agent reports malicious code snippets with explanations. ProfMalPlus achieves a 98.1% F1-score, outperforming state-of-the-art detectors by 3.5% to 52.6%. It also identified 597 previously unknown malicious packages, all confirmed and removed from NPM.
arXiv:2607.13967v1 Announce Type: new
Abstract: Adapting a multilingual encoder to a new language \emph{and} a new task with only a few hundred gold examples is a common low-resource NLP setting, yet the two axes are usually fused via an expensive language--task fine-tuning run. We ask whether they can instead be trained separately and recombined in weight space. \DeltaMergeLowRes{} learns a language delta $\Delta_L$ from unlabeled monolingual text and a task delta $\Delta_T$ from labeled English data, then composes them at inference under one of four rules: additive, activation-guided, sparsity-aware, and a novel \emph{cross-axis TIES}. The new rule adapts the TIES-Merging steps of trimming, sign election, and merging to the language and task axes rather than to two task axes. Holding $(\Delta_L,\Delta_T)$ fixed across rules on four task families and four African languages ($158$ evaluated cells, $10{,}000$-sample paired bootstrap per cell), we find: (i) cross-axis TIES wins summarisation on $3/4$ languages by $+4$ to $+7$ chrF (chrF $18.59$ vs.\ $13.80$ task-only); (ii) it improves QA F1 by $+2.32$ and EM by $+2.91$; and (iii) sparsity-aware merging cuts classification ECE by $36\%$ at parity macro-F1. The composition rule materially changes what the merged model preserves, suppresses, and calibrates. We release all JSON traces and a claim ledger.
arXiv:2607.13978v1 Announce Type: new
Abstract: Music-driven dance generation aims to produce human motion that is both rhythmically synchronized and semantically consistent with music. While recent neural approaches have achieved impressive visual realism, they typically model motion as a continuous signal and neglect its compositional nature, making generated dances structurally incoherent and difficult to control. In this work, we introduce a structure-aware framework that models choreography as a sequence of atomic movements-semantically interpretable motion events that serve as the building blocks of dance. To construct this atomic movement vocabulary, we first segment large-scale dance data and cluster them into atomic movement groups. We then employ a large language model to semantically relabel and refine the clusters, yielding a set of interpretable and reusable atomic movements. Based on these atomic movement annotations, we design a two-stage generation framework that mirrors the human choreography process. In the atomic movement planning stage, the model predicts the type, duration, and timing of atomic movements conditioned on the input music, forming a symbolic dance allocation. In the completion stage, a transition-aware generator synthesizes smooth and stylistically coherent motion conditioned on the planned structure. Extensive experiments demonstrate that our method produces dances with significantly improved structural coherence, rhythmic alignment, and perceptual naturalness compared to existing baselines, while providing enhanced interpretability and controllable editing through explicit structural representation.
arXiv:2607.13285v1 Announce Type: new
Abstract: The capability of a modern AI agent depends not only on its foundation model but also on its harness, which constructs prompts, manages state, invokes tools, and coordinates execution. As models, APIs, environments, and requirements evolve, the harness must be continually modified. Before such a change can be made, a developer or coding agent must identify all code locations that implement the target behavior. This is difficult because production harnesses are large, tightly coupled, and behaviorally distributed, while modification requests describe what the system should do and repositories are organized by files and modules. Code search, repository indexing, and long-context processing ease inspection, but still leave this behavior-to-code mapping to be recovered by hand. Behavior localization is therefore a central bottleneck in harness evolution. We introduce the Harness Handbook, a behavior-centric representation synthesized automatically from a harness codebase via static analysis and LLM-assisted structuring, linking each behavior to its corresponding source. We also introduce Behavior-Guided Progressive Disclosure (BGPD), which guides agents from high-level behaviors to relevant implementation details and verifies candidate locations against the current source. On diverse modification requests from two open-source harnesses, Handbook-Assisted planning improves behavior localization and edit-plan quality while using fewer planner tokens, with the largest gains on scattered sites, rarely executed paths, and cross-module interactions. Evolving complex agentic systems thus depends not only on generating edits, but also on determining where those edits should be made.
arXiv:2607.13983v1 Announce Type: new
Abstract: Vision Transformer (ViT) has been widely used as a powerful framework for modeling global dependencies among image patches. However, its core component, self-attention assigns softmax-normalized relative weights to all patches, making it difficult to evaluate the relevance between patches independently. In visual recognition, images often contain many background or redundant patches, yet self-attention cannot explicitly reject such irrelevant patches, which may introduce unnecessary information into feature aggregation. To address this limitation, Screening has been proposed in the field of language modeling, where the relevance of each token is independently evaluated based on query-key similarity and low-relevance tokens are explicitly excluded through thresholding. In this work, we propose VisionScreen, a new vision model that extends Screening mechanism to visual recognition. VisionScreen treats image patches as tokens arranged on a two-dimensional grid and extends absolute relevance estimation based on query-key similarity to the two-dimensional spatial domain. This allows each patch to selectively aggregate only content-wise and spatially relevant patches without relying on competition among patches. Experiments on image classification benchmarks demonstrate that the proposed method outperforms conventional ViT. These results suggest that Screening can be effective for visual recognition, offering an alternative to relative feature aggregation based on softmax attention.
arXiv:2607.13284v1 Announce Type: new
Abstract: The minimum spanning tree (MST) problem is one of the most basic optimization problems on metric spaces and graphs. We study the problem of computing a $(1+\epsilon)$-approximation to the MST of an $n$-point metric space $(X, \mathbf{d})$ of doubling dimension $\mathrm{ddim}$. In doubling metrics, previous deterministic algorithms incur a running time with dependence $\epsilon^{-O(\mathrm{ddim})}$.
We give a deterministic algorithm that computes a $(1+\epsilon)$-approximation to MST in time $2^{O(\mathrm{ddim})} n \bigl(\log n + \epsilon^{-1} \log^4(1/\epsilon)\bigr)$. For bounded doubling dimension, this improves the previous dependence on $\epsilon$ from $\epsilon^{-O(\mathrm{ddim})}$ to essentially linear in $\epsilon^{-1}$. Moreover, as a special case, our result improves the previous best deterministic running time for bounded-dimensional Euclidean metrics due to Arya and Mount~[SODA'16] by almost a factor of $\epsilon^{-1}$. We also show that, unlike in bounded-dimensional Euclidean spaces, MSTs in bounded doubling metrics can have arbitrarily large maximum degree, while every doubling metric nevertheless admits a $(1+\epsilon)$-approximate MST of maximum degree $2^{O(\mathrm{ddim})}\log(1/\epsilon)$.
arXiv:2607.13585v1 Announce Type: new
Abstract: Compact deuterium--deuterium (D--D) neutron sources can provide controllable irradiation fields for lithium-blanket studies, although their broad joint energy and angle distributions differ from the conventional $2.45$~MeV isotropic representation. We couple particle-in-cell (PIC) simulations of target-normal-sheath-accelerated deuterons with a thick-target $D(d,n)^{3}$He source model and Monte Carlo neutron transport. For natural lithium, the seven two-dimensional sources change tritium production per source neutron by $-2.5\%$ to $+54.1\%$ relative to the ideal source. The matched three-dimensional calculation gives an increase of $43.5\%$ and lowers the corresponding ratio from $1.5406$ to $1.4350$. Source substitutions show that the difference is predominantly spectral, since the real spectrum alone gives a factor of $1.4199$, while using the real neutron emission directions in place of isotropic emission adds only a further factor of $1.0106$ in the three-dimensional case. The real spectrum lowers the $^{6}$Li contribution by $6.9\%$, but the accessible $^{7}$Li$(n,Xt)$ response exceeds this loss. Enrichment to $90\%$ $^{6}$Li keeps the total change within $\pm1.5\%$. In the matched three-dimensional converter and blanket calculation, direct $D(d,p)$T production is $0.8458$ tritons per source neutron and accounts for $98.1\%$ and $86.9\%$ of the combined production for natural and enriched lithium, respectively. High-density polyethylene moderation raises tritium production by about one order of magnitude but first weakens and then reverses the increase in blanket tritium production. The analysis quantifies source-model effects in compact breeding tests.
arXiv:2607.13986v1 Announce Type: new
Abstract: The 11th Affective Behavior Analysis in-the-wild (ABAW11) Multi-Task Learning Challenge requires a unified system to predict valence-arousal, categorical expressions, and facial action units from the official s-Aff-Wild2 images. Although these tasks are naturally related through facial behavior, our validation experiments show that they benefit from different visual features, temporal processing strategies, fusion mechanisms, and calibration procedures. In this paper, we study task-adaptive feature fusion for ABAW11 multi-task affective behavior analysis. We first adapt two pretrained visual backbones, DINOv2 ViT-L and DINOv3 ConvNeXt-base, on an external expression-oriented facial image set and then freeze them to extract complementary frame-level features from the official ABAW11 data. On top of these frozen features, we systematically compare frame-level prediction heads, temporal convolutional heads, post-hoc temporal smoothing, LightGBM models, feature concatenation, gated fusion, residual fusion, late logit fusion, threshold calibration, and shared MTL structures. The final system selects task-specific fusion and prediction strategies rather than forcing all tasks to share a single architecture. On the ABAW11 validation set, the selected system achieves an EXPR macro-F1 of 0.4222, an AU macro-F1 of 0.5402, and a mean VA CCC of 0.6717, resulting in an overall validation score of 1.6341. The results suggest that task-adaptive fusion of frozen visual features is a simple and effective strategy for ABAW-style multi-task affective behavior analysis.
arXiv:2607.13963v1 Announce Type: new
Abstract: Strip scintillator detectors are used nowadays in many fields of applied physics, particularly in medicine, civil engineering, mapping of underground resources, and security. In this work we provide an analytical description of light transport in a cuboid-shaped strip scintillator detector from the scintillation location to the detecting surface. We use kaleidoscopic-ray-tracing approach to reproduce the average time profile of the energy deposition (light collection) in the detecting surface. We demonstrate the applicability of the model on the case of PWO crystal of $1.5\times 1.5 \times 30$ cm$^3$ size. We show that the results achieved in the model are in good agreement with Monte Carlo (MC) simulation. Notably, the developed model requires dozens of milliseconds to be implemented, thus, it is applicable for real-time calibration of the detector, while the MC simulation takes hours. Also, we highlight that the kaleidoscopic-ray-tracing itself being used in MC simulations can highly speed them up in the case of specular reflectors.
arXiv:2607.13972v1 Announce Type: new
Abstract: While intrasaccular flow disruptors are widely used to treat wide-neck intracranial aneurysms, state-of-the-art patient-specific computational models routinely neglect the deployment mechanics by prescribing a pre-seated geometry. This shortcut oversimplifies the true physics and misrepresents the Contour Neurovascular System (CNS), whose critical biomechanical features, such as neck coverage, wall apposition, and migration resistance, are highly path-dependent. To resolve this limitation, we present a contact-resolved finite-element framework that explicitly computes the structural mechanics of implant deployment within patient-specific vascular environments. The device is discretized as a dual-layer interwoven Nitinol braid using geometrically exact beams, while the vascular wall is represented as a deformable hyperelastic shell. Non-linear frictional contact formulations govern complex wire-wire and wire-wall interactions under a staged release protocol. Evaluating three anatomical phenotypes reveals that the final equilibrium morphology is highly sensitive to tangential slip resistance and vertical release depth. Frictionless assumptions permit excessive post-contact sliding, whereas near-stick conditions enhance anchoring but restrict local compliance. Crucially, conventional geometric fast placement fails to capture these critical contact interactions and wall-supported mechanical equilibrium. This deployment-resolved framework establishes a biomechanically grounded foundation for downstream hemodynamics, fluid-structure interaction, and mechanobiological thrombus-formation modeling.
arXiv:2607.13331v1 Announce Type: new
Abstract: Retail demand forecasts are reused across replenishment, capacity, labor, and transportation planning cycles. Point-error objectives do not constrain abrupt movement between adjacent forecasts, while post-hoc smoothing acts only after model fitting. We ask whether a training-time penalty on consecutive within-series movement can improve horizontal forecast-path stability without materially changing point accuracy. The penalty is evaluated in a temporal-structured pipeline combining recent-demand embeddings with calendar, price, hierarchy, item, and store features. On selected M5 demand series at 1000, 3000, and 4000-series scales, the stability-aware hybrid model improves Forecast Stability Score over XGBoost by 6.91%, 6.66%, and 7.68%, respectively, while RMSE changes remain within 0.72% across three random seeds. Post-hoc exponential smoothing attains lower raw movement but incurs a larger RMSE cost; training-time regularization preserves more point accuracy and performs favorably under normalized stability. These findings extend forecast evaluation from point-error minimization toward an accuracy-stability trade-off perspective for operational retail forecasting.
arXiv:2607.13988v1 Announce Type: new
Abstract: Multi-turn agents solve complex tasks through extended sequences of tool interactions before producing a final answer, making credit assignment a fundamental challenge during post-training. Outcome rewards provide reliable supervision for short-horizon reasoning, but become sparse and high-variance as trajectories grow to tens or hundreds of tool calls. They can also be misleading: a failed rollout may contain many useful actions that move the agent closer to the goal, yet outcome-only training assigns them the same negative advantage as the eventual mistake. We propose TRACE (Turn-level Reward Assignment via Credit Estimation), a dense credit-assignment method for agentic reinforcement learning. TRACE represents rollouts as state transitions at tool-call boundaries, obtains gold-answer log-probabilities from a frozen reference model, transforms them into log-ratio state values, and derives per-action rewards as Temporal-Difference changes in those values. This requires no additional critic or process-label training, and its one-step log-ratio TD component telescopes across redundant tool calls. On long-horizon complex search, TRACE substantially improves base-model tool-use ability using pure RL, without a cold-start supervised fine-tuning stage, an agentic mid-training stage, or training on live-web data. On the closed-web BrowseComp-Plus benchmark, it raises Qwen3-4B from $7.2$ to $35.6$ and Qwen3-30B-A3B from $8.4$ to $42.6$. The learned search behavior also transfers to open-web benchmarks, and the learning curves show earlier improvement and faster convergence during RL training.
arXiv:2607.14018v1 Announce Type: new
Abstract: We investigate how each component of the Transformer feedforward block architecture design determines how much rank survives across depth at initialization. We reinterpret skip connections and normalization, long understood as controlling magnitude, as mechanisms for preserving gradient rank across depth, since the very matrix multiplications and nonlinear activations that make the network expressive also reduce the rank. We show that skip connections trade off rank collapse against ensemble-like behavior, controlled by the relative scales of the branch and the skip: skip connections route the gradient around the residual branch, where rank is lost, rather than along the long gradient paths that encourage the layers to compose. The placement of the normalization layer controls this same tradeoff by setting the branch-to-skip ratio across depth, unifying much of the normalization placement and depth scaling literature, in particular why rank collapses for Post-Norm but plateaus for Pre-Norm. Other aspects of the architecture, like the two-matrix structure that expands and contracts the width, use additional parameters to preserve the representation or branch Jacobian rank. The second matrix decorrelates a coherent mean spike that would grow across blocks with a single matrix and uncentered activation, preventing the residual representation from collapsing. The width expansion between the two matrices keeps the branch Jacobian full rank: applying the rank-reducing activation in this expanded space leaves enough directions to span the original, at a width that follows a Marchenko--Pastur law. The initialization rank of the input--output Jacobian predicts which networks train on CIFAR-10. Taken together, we recast architecture design for deep networks as navigating an intrinsic tradeoff among rank collapse, ensemble-like behavior, and parameter count.
arXiv:2509.15520v2 Announce Type: replace
Abstract: Experimental evidence has shown the stability of oxygen-stress-damaged guanine, known as 8-oxoguanine. This common oxygen-damaged nucleobase is often found in the presence of reactive oxygen species and can result in the mispairing between adenine and 8-oxoguanine in a Hoogsteen pair. We have computationally investigated the role of 8-oxoguanine to support experimental results and focus the investigation towards the polymerase structure, identifying how 8-oxoguanine interacts in the polymerase environment. Quantum mechanical investigations show the Hoogsteen pairing of adenine and 8-oxoguanine is the most energetically favourable state compared to a Watson-Crick state, supporting experimental evidence. Molecular Dynamical calculation of 8-oxoguanine located in B-DNA provide an average C1' backbone spacing of 1.1 nm compared to adenine-thymine spacing of 1.05 nm but remains within the error margin, however when in the polymerase environment, 8-oxoguanine and the canonical adenine - thymine spacing match at 1.11-1.12nm spacing while 8-oxoguanine is located on the template strand, but is disrupted by tyrosine 251 while situated on the triphosphate strand. We observe that 8-oxoguanine, when paired with adenine in polymerase {\lambda}, can mimic the adenine-thymine structure, as predicted in experimental results, and thus would be misincorporated, but is strongly dependent on its conformation.
arXiv:2607.13955v1 Announce Type: new
Abstract: We study the online proportional knapsack problem with two paid forms of recourse. Items arrive one by one and must be handled immediately, without knowledge of the future: an algorithm may pack an item $x$, reject it, or reserve it for possible later use at proportional cost $\alpha x$; additionally, it may at any time remove previously packed items, at proportional cost $\beta y$ for each removed item $y$. Reservation and removal have each been analyzed in isolation, but their combination raises a natural question: is the better of the two mechanisms always optimal on its own, or is there a region in the parameter space spanned by $\alpha$ and $\beta$ in which they genuinely enter into a symbiosis? So far, this question has only been answered for the special case of free removal ($\beta = 0$), leaving the vast majority of the parameter space unexplored.
We close this gap, determining matching upper and lower bounds on the competitive ratio for every pair of cost parameters $(\alpha, \beta)$ and revealing three qualitatively different regimes. In some regions, reservation alone already achieves the optimal ratio; in others, removal alone does. However, most interestingly, in the heart of the parameter space lies a symbiosis region in which combining both mechanisms is strictly better than either one on its own. The optimal algorithm in the symbiosis region is a natural blend of the two known single-mechanism strategies: postponing commitment by reserving until a threshold is reached, then packing greedily and revising via removal.
arXiv:2607.13977v1 Announce Type: new
Abstract: Aspect-Based Sentiment Analysis (ABSA) requires models to identify sentiment toward specific aspects rather than relying on the global polarity of a sentence. This makes counterfactual evaluation especially challenging: a valid counterfactual should flip the sentiment of one target aspect while preserving the sentiment of all non-target aspects, semantic meaning, fluency, and factual consistency. Existing counterfactual generation methods often focus on sentence-level label flipping and may produce edits that are fluent but aspect-invalid, semantically drifting, or contradictory. To address this limitation, we propose CAVE-ABSA, a Constraint-Aware Validated Editing framework for generating and validating aspect-level counterfactuals. CAVE-ABSA localizes the opinion span associated with the target aspect, performs controlled counterfactual rewriting, refines candidates through a repair module, and filters them using aspect-level verification, semantic similarity, AMR-guided structural preservation, edit minimality, fluency, and contradiction detection. The framework is designed to construct validated counterfactual ABSA datasets for robustness evaluation and data augmentation. By explicitly separating generation from validation, CAVE-ABSA provides a principled approach for producing meaningful aspect-local counterfactuals and for testing whether ABSA models truly rely on aspect-grounded sentiment reasoning.
arXiv:2607.13324v1 Announce Type: new
Abstract: The existence, several properties, and constructions of Generalized Weighing-Hadamard (GWH) matrices over finite fields are addressed in this work. We study the subset of invertible GWH matrices and show that it forms a group under matrix multiplication. Besides that, we introduce a strong notion of equivalence between such matrices, defined via orthogonal transformations, and further prove that the corresponding quotient group by the subgroup of orthogonal matrices is abelian. Finally, we discuss some applications of these matrices in coding theory
arXiv:2607.13350v1 Announce Type: new
Abstract: The power of Large Language Models (LLMs) has led us to investigate how they might be fine-tuned for learning the "language of molecular geometry". The fine-tuning of the LLMs using Cartesian coordinates or Z-matrices provides an extremely simple method for accurately predicting equilibrium structures and diverse sets of conformers of small organic and drug-like molecules with excellent accuracy and outperforming specialized deep learning models. While the most common representation of molecular geometries are Cartesian coordinates performs adequately, we find that the inherent invariances and relational nature of geometries represented as Z-matrices provides a better grammar for LLM adaptation. Finally, we show that enhancing an LLMs capabilities for robust prediction of small molecule geometries still retains nearly all of its pre-trained language abilities by randomly mixing in small quantities of natural language prompt-response pairs into the fine-tuning.