Papers

We study logistic regression on linearly separable data under gradient descent with a large constant stepsize $\eta$. Such dynamics may exhibit a characteristic Edge of Stability phenomenon, in which the loss initially oscillates before transitioning to a stable phase of monotone decrease. Existing work provides a tight $\Theta(1)$ bound in dimension $d=2$ as $\eta \to \infty$ and conjectures a bound independent of $\eta$ in arbitrary dimensions $d\geq 2$. In this paper, we disprove this conjecture by showing that, for every fixed sample size $n\geq 2$ and sufficiently small margin $\gamma$, the worst-case transition time is $$\Theta\!\left((\log\eta)^{\min\{n-2,d-2\}}\right)$$ uniformly over $d\geq2$. The key challenge in establishing a tight bound is that the sample contributing most strongly to the gradient can change repeatedly across iterations. To address this issue, we control such changes by induction on dimension and sample size, and construct matching hard instances.

Verifier-free reinforcement learning with probability-based rewards offers a promising way to train LLMs on general reasoning tasks where external verifiers are unavailable. Yet the reliability of these rewards, especially in long-horizon reasoning, remains underexplored. This work identifies a length-dependent failure mode of probability rewards, which we call the Posterior Concentration Phenomenon (PCP). We show that the probability of a reference answer conditioned on a reasoning trace often collapses to a low-variance interval as the trace becomes lengthy. This phenomenon results in nearly indistinguishable rewards, which, under GRPO-based settings, makes probability-based policy optimization unstable and inefficient. Motivated by this, we propose Reinforcement Learning with Concentration-aware Posterior Rewards (RLCPR), a verifier-free RL framework to explicitly account for PCP for better optimization stability and token efficiency. It has two components: uncertainty-aware data sampling, which reduces concentration-prone rollouts before generation, and concentration-aware regularization, which penalizes unnecessarily long traces when posterior rewards collapse. Extensive experiments show that, alongside higher token efficiency, RLCPR outperforms the state-of-the-art verifier-free RL baseline by up to 4.0% on six of seven benchmarks, including general-domain and mathematical reasoning challenges.

We present a phenomenological study of a hypothetical compact object with a traversable 2 Kerr-type wormhole geometry connecting two asymptotically distinct regions. One mouth of the wormhole is assumed to reside in an active galactic nucleus (AGN) environment with ongoing accretion, while the second mouth is located in a comparatively quiescent galactic nucleus without accretion activity. In such a configuration, matter accreted near the AGN-side mouth may traverse the wormhole throat and emerge from the opposite mouth. Because the inflowing matter possesses angular momentum, the resulting outflow is expected to be predominantly equatorial, producing a disk-like structure fundamentally different from a conventional accretion disk around a black hole. Several dynamical regimes are considered depending on the outflow velocity relative to the local escape and orbital velocities. Possible observational manifestations, thermal properties, recombination signatures, and magnetic field effects are discussed. Special attention is given to the possibility of jet formation in the absence of a conventional accretion disk. We explicitly distinguish the weak-field, large-radius Newtonian classification used for the outer flow from the strong-field region near the throat, where the dynamics should be treated in a general-relativistic and magnetohydrodynamic framework. We further discuss the specific energy and angular momentum inherited from the inner edge of the accretion flow, the possible modification of these quantities by magnetic stresses and energy-extraction processes, and the effects of an intrinsic dipolar magnetic field of the wormhole.

Given a set of $n$ points $X$ in a metric space and an integer $k$, max-min diversification aims to select $k$ points of $X$ maximizing their minimum pairwise distance. This objective function is however highly vulnerable to noisy points. In[Amagata, AAAI23], a robust formulation is proposed which addresses this vulnerability by excluding solutions containing any of $z$ outliers, defined as the $z$ points in $X$ with the largest nearest-neighbor distances. That paper also presents a coreset-based streaming algorithm for the new formulation, based on a suitable inlier-outlier separation assumption. However, we identify three shortcomings in the algorithm by [Amagata, AAAI23]: its coreset construction requires an offline computation over $X$, which needs memory linear in $n$, in stark contrast with the typical goals of stream processing; the one-pass procedure used to extract the solution from the coreset may return fewer than $k$ points (hence, an unfeasible solution) because it permanently discards points too far from the current solution; and its outlier-exclusion guarantee is only probabilistic and weakens as the coreset size shrinks. In contrast, we present a deterministic coreset-based algorithm that, under a natural inlier-outlier separation assumption (similar to the one used in [Amagata, AAAI23]), returns exactly $k$ inliers which are a $(2+\varepsilon)$-approximate solution, for any $\varepsilon>0$, thus only $\varepsilon$ above the best polynomial-time sequential approximation, even without outliers. Its one-pass streaming implementation adapts obliviously to the dataset's doubling dimension $D$ and, for wide ranges of $k$, $z$, $\varepsilon$, and $D$, it uses memory independent of $n$. For sufficiently long streams, its amortized update time is proportional to the coreset size, thus also independent of $n$.

We study how driving history affects the nonlinear topological relaxation of a holographic superfluid. Different histories are constructed to reach the same present driving conditions and then follow the same future protocol. Although they relax to the same final winding sector, their late-time dynamics remain strongly history-dependent: the residence time of the penultimate winding sector changes by about a factor of \(3.7\). This variation is organised by the modal imbalance carried into the common amplification stage. Histories with different shapes but matched modal imbalance produce nearly identical relaxation times. The same qualitative behaviour is found for a second pair of competing modes.

Extended Lya emission is routinely detected around high-redshift quasars, revealing large reservoirs of cool circumgalactic medium (CGM). Its resonant nature complicates interpretation of gas density, ionization state, and kinematics, making non-resonant tracers such as HeII and Ha essential. Extending previous analytical models, we use CLOUDY photo-ionization simulations to predict HeII/Ha in quasar-illuminated gas over a broad parameter range, including log-normal density distributions. The ratio constrains density fluctuations, commonly quantified by the clumping factor C. We compare the models with extended Lya, HeII, and Ha observations around quasars at z~2.1-4.5. At z~2.2, we observed 13 quasars with Keck/KCWI and targeted the systems with the most prominent extended Lya and HeII emission with Keck/MOSFIRE, providing the first statistical quasar-CGM sample with combined Lya, HeII, and Ha measurements. Within matched apertures out to ~35 pkpc, stacked spectra give Lya/Ha=9.1, HeII/Ha=0.54, and HeII/Lya=0.06. Single-density models require very high CGM densities (n>10 cm^-3) or implausibly soft AGN spectra to reproduce the low HeII/Ha ratio. Broad log-normal density distributions instead reproduce it at lower median densities but require a substantial emissivity-weighted high-density tail, corresponding formally to C~10^3-10^4 in our parameterization. We also analyze 26 VLT/MUSE quasars at 3.1<z<4.5, where Ha is inaccessible from the ground, finding 14 systems with co-spatial extended Lya and HeII emission. The detected, emission-weighted HeII/Lya component shows no evidence for strong evolution between z~2.2 and z~3.7, though this does not imply an invariant underlying density PDF. These constraints provide a basis for comparison with theoretical models and hydrodynamical simulations of the physical processes shaping quasar CGM at high redshift.

Deep neural networks are increasingly deployed in long-lived systems, where task requirements may change after training. In such settings, part of the original output space may become obsolete: a class, prediction region, or learned behaviour may no longer be valid. Existing approaches either leave the obsolete behaviour intact or require fine-tuning, which can be expensive. We propose Deep Repurposing (DR), a post-hoc framework for adapting models under task obsolescence. DR estimates the latent geometry of obsolete and retained regions, removes obsolete-supporting components, and reallocates retained-compatible evidence through an analytic repair map without gradient updates. This yields repaired predictions and representations in which obsolete regions no longer act as valid outputs, while useful obsolete structure can support the retained task. Across multiple task settings, DR removes obsolete behaviour while preserving retained utility. More importantly, across classification benchmarks, DR matches or exceeds competing unlearning and editing baselines in retained accuracy, eliminates obsolete predictions, and adapts up to $60\times$ faster than competing unlearning methods.

While state-of-the-art automated models for medical image segmentation achieve high mean performance, they frequently suffer from localized, catastrophic failures that preclude safe clinical deployment, particularly in neuro-oncology. Interactive segmentation frameworks mitigate this by incorporating human oversight, but traditionally impose prohibitive cognitive and temporal workloads by requiring clinicians to manually search for errors. In this project, we present an efficient, Hybrid Structural-Aleatoric Human-in-the-Loop framework for glioma segmentation that bridges the gap between automated baseline performance and surgical-grade precision, achieving sub-2.0 mm HD95 on curated benchmarks while providing safety-net routing for structural failures across real-world clinical data. By extracting voxel-wise Test-Time Augmentation (TTA) uncertainty and applying hierarchical topological filtering, our method proactively isolates high-risk structural anomalies. We comprehensively evaluated our approach on a challenging out-of-distribution clinical stress-test cohort (N = 362). Operating under a simulated Human Oracle, the framework improved the Whole Tumor (WT) Dice score from 0.891 to 0.914 and reduced the 95th percentile Hausdorff Distance (HD95) from 5.82 mm to 4.76 mm. Critically for surgical safety, the system rescued severe boundary failures in the Tumor Core, reducing mean HD95 from 17.96 mm to 14.83 mm (improving absolute TC Dice to 0.356). These spatial rescues were achieved while demanding a median interactive workload of just 11.3% of the target volume. Acknowledging this as a simulated upper bound lacking real-world cognitive friction, the framework nevertheless demonstrates a highly Pareto-efficient pathway for safely deploying clinical AI.

Personalized interpretation of health checkup results requires reasoning across longitudinal records, medical knowledge, lifestyle guidance, and healthcare navigation. We present a multi-agent large language model (LLM) system that identifies multiple intents, maps each to a task-specific agent, executes them in parallel, and synthesizes their outputs. We compared answers generated in Single Agent and Multi Agent settings on 120 Korean compound queries combining two to four requirements, using synthetic health checkup records. The Multi Agent improved the weighted LLM-judge score from 1.695 to 1.797 (p = 0.027), and three additional LLM judges showed consistent improvements ($\Delta$ = +0.111 to +0.186, all p < 0.05). The gains came from usefulness, consistency, and the handling of every requirement in compound queries, whereas numerical accuracy and grounding improved significantly under only one of the four judges and medical safety did not differ, and critical failures occurred at similar rates (Single Agent 15.0% vs. Multi Agent 13.3%). Two human evaluators preferred Multi Agent in 66.7% and 68.3% of pairwise comparisons. Multi Agent execution increased latency and cost by 1.31$\times$ and 2.02$\times$, respectively. In exploratory subgroup analyses, the improvement was concentrated in queries involving personal-record lookup.

Code-switched (CS) speech leaks through the monolingual language identification (LID) filters used to curate massive speech corpora, calling for CS-aware LID (CS-LID). We formulate utterance-level CS-LID as multi-label language-set prediction and propose a set generator that directly outputs the languages in an utterance, comparing it against atomic-pair and score-based classification baselines. Oracle Top-k is the strongest baseline, but thresholding fails because no single threshold separates CS from monolingual speech. Our set generator predicts the correct language count on unseen pairs without assuming the number of languages, but underperforms oracle Top-k in exact set accuracy. Our analysis identifies the key obstacles to robust CS-LID: oracle cardinality, threshold instability, language bias in CS training data, and the synthetic-to-real gap.

This letter introduces an optimization-based fixed time control algorithm for solving the voltage regulation problem of a radial and balanced power distribution network. The proposed algorithm requires no prior knowledge of the exact network impedance (resistance and reactance), yet guarantees voltage convergence to the predefined safe limit within a fixed time-window. We embrace results from Fixed-time stability (FxTs) and Control Lyapunov function (CLF) to analyze the stability and robustness of the underlying algorithm and then synthesize it leveraging the Quadratic Programming (QP) approach. We first analytically provide sufficient conditions on the control gains and the design parameters, that can ensure voltage regulation in fixed-time. Thereafter we transform the existing voltage regulation problem into an equivalent QP-based optimization framework and translate the aforesaid conditions to find feasible solutions for voltage stability. We empirically verify the efficacy of our proposed algorithm on the IEEE-33 bus distribution network and compare its performance against other existing robust control methods.

Skew-symmetric probability distributions provide a principled mechanism for incorporating gradient information into Markov chain Monte Carlo algorithms. Here we review the (preconditioned) Barker proposal, a Metropolis–Hastings algorithm built on skew-symmetric distributions, and motivate its design. We then introduce three natural extensions. First, we propose coordinate-free variants of the Barker algorithm. Second, we introduce a Gibbs-style Barker algorithm that re-evaluates the gradient at each partially updated coordinate. Third, we derive a simplified manifold Barker algorithm, producing a manifold sampler with enhanced robustness compared to natural comparators. Numerical experiments demonstrate that the Gibbs-style variant improves raw sampling efficiency on correlated targets, that the coordinate-free variants offer limited practical advantage over the standard Barker proposal once computational costs are accounted for, and that the simplified manifold Barker algorithm can achieve significant advantages over simplified manifold MALA when the local geometric structure of the target is irregular or unreliable.

The Kannan–Lovász–Simonovits (KLS) conjecture asks whether every isotropic log-concave probability measure on $\mathbb R^n$ has a Cheeger constant bounded below by a universal positive constant. We prove that $\psi_n\le C$ for a universal constant $C>0$, resolving the KLS conjecture. We also prove that $C_P(\mu)\le C'$ for every isotropic log-concave probability measure $\mu$ on $\mathbb R^n$, with a universal constant $C'>0$.

Barren plateaus hinder training of parameterized quantum circuits by making loss gradients exponentially small in the number of qubits. For angle-encoded product states, we show how the input distribution affects output variation through algebraic input purity, i.e. the input's overlap with the circuit's dynamical Lie algebra. With the specified readouts and Haar or exact group 2-design sampling, matchgate circuits on $n$ qubits retain output variance of order $1/n$ for every pure product input. The off-diagonal family instead has zero output on computational-basis inputs at any depth and for every parameter choice. Independent uniform angles yield mean variance of the same order as matchgates. A count of diagonal Pauli strings identifies these zero-output endpoints. For any fixed dataset of nonzero inputs with at least $2n$ coordinates, we construct a classical preprocessing certificate. A shared random rotation is accepted only when the dataset's mean purity passes a computable threshold. This certifies mean output variance of order $1/n$ for the off-diagonal family before circuit execution, with a constant expected number of trials. Binary and ternary weighted encodings also recover the independent-angle mean purity from one uniform scalar input. Numerical experiments examine training behavior and extensions to larger algebras. The guarantee concerns output variance averaged over inputs, while successful learning also depends on the task and information preserved by the encoding.

UMAP achieves scalable layout optimization through stochastic negative sampling. However, this stochasticity can lead to unstable embeddings across reruns and downstream reuse, as the estimated repulsive forces depend on the ordering of sampling events. We present ibUMAP, a coherent field-based alternative that evaluates attraction and repulsion from a shared embedding snapshot and applies them synchronously. Its degree-weighted repulsive field is motivated by the conditional expectation of negative sampling for a fixed embedding and represented by three scalar moments, which are evaluated efficiently on CPUs and GPUs using an interpolation-based FFT scheme. This formulation avoids explicit all-pairs computations while inducing optimization dynamics that differ from those of standard online UMAP. Controlled experiments show that synchrony and kernel capping alter the local-global fidelity trade-off, whereas FFT evaluation produces small average changes in final quality. End-to-end benchmarks show median speedups of 3.29x unseeded and 5.79x seeded over umap-learn on CPU, and 1.44x over cuML on million-scale datasets under unseeded GPU execution. These gains accompany greater run-to-run stability and measurable fidelity trade-offs.

Context. Ultraluminous X-ray sources (ULXs) are thought to be powered, in many cases, by super-Eddington accretion onto compact objects. While soft X-ray lags have been detected in several ULXs, hard lags remain rare and poorly understood. Aims. We investigate the temporal and energy dependence of X-ray lags in NGC 7456 ULX-1 to constrain their physical origin and probe the super-Eddington accretion flow. Methods. We analyzed the two deepest XMM-Newton observations, taken in 2018 and 2023. Hard (1-10 keV) and soft (0.3-1 keV) light curves were cross-correlated using adaptive-binning techniques optimized for Poissonian low-count data. We measured lags over consecutive 10 ks intervals and investigated their energy dependence. Spectra were modeled using thermal and Comptonization models. Results. We detect significant hard X-ray lags in both observations, with the hard emission delayed by $\sim 10^3$ s during phases of rapid flux variability. The delays are primarily driven by the lowest-energy photons. Spectral modeling indicates a Comptonization-dominated flow comprising a cooler, extended outer region and a hotter, compact inner flow embedded in an optically thick wind. We interpret the delays as the combined effect of inward propagation of accretion-rate fluctuations and photon diffusion within the dense outflow. Fluctuations first enhance the soft-emitting outer regions and then propagate toward the hotter inner flow, where photons undergo stronger Comptonization before escaping with a kilosecond delay. The small inferred inner emitting radius disfavors an intermediate-mass black hole accretor. Conclusions. The sign, amplitude, and energy dependence of the delays disfavor standard reverberation. Propagation-driven variability coupled with radiative transfer in optically thick winds appears to play a major role in shaping the timing properties of super-Eddington accretion flows.

We present GridSMR, a sharded blockchain that scales execution horizontally while allowing dependent cross-shard operations to progress within a single block. Existing sharded systems typically place coordination between dependent cross-shard steps, making latency grow with causal depth. GridSMR localizes atomicity to individual accounts and executes cross-account work asynchronously. Using an execute-before-agree architecture, dependent operations execute across shards as they become available, while consensus later validates and commits the resulting schedule. This enables Causal Compression: cross-shard latency need not grow with the causal depth of a computation. GridSMR scales single-validator execution to 1.07M requests/s and four-validator execution to 193K committed requests/s, while reducing 16-hop causal-chain latency by 8.2x versus deferred execution.

We continue the study of $k$-limited domination in graphs, a domination variant in which each vertex of a dominating set may dominate at most $k$ vertices outside the set. This concept extends classical domination by incorporating capacity constraints on dominating vertices. We improve the general lower bound for the $k$-limited domination number $\gamma_k^L(G)$ by employing the concept of $k$-capacitated domination. As a consequence, we characterize graphs satisfying $\gamma_k^L(G)=\lceil \frac{n}{k+1} \rceil$. We also refine the known upper bound for graphs with $k<\delta(G)$ using the $k$-limited packing number. Under this condition, we describe graphs attaining $\gamma_k^L(G)=n-k$. These results extend and unify previous investigations for the case $k=1$ and provide a complete characterization of graphs attaining the extreme values of the $k$-limited domination number under the considered assumptions.

A strong odd $k$-coloring of a graph $G$ is a proper $k$-coloring such that every color appearing in the neighborhood of a non-isolated vertex appears an odd number of times. The minimum $k$ for which $G$ admits a strong odd $k$-coloring is the strong odd chromatic number, denoted by $\chi_{\text{so}}(G)$, of $G$. Given a graph $G$ and an integer $k$, \textsc{strong odd $k$-colorability} problem asks whether $G$ admits a strong odd $k$-coloring. It is known that STRONG ODD $k$-COLORABILITY is NP-complete in general graphs. In this paper, we prove that the problem is NP-complete on perfect elimination bipartite graphs for $k\geq3$, which is a subclass of bipartite graphs. Furthermore, we show that $\chi_{\text{so}}(G)$ is inapproximable within a factor of $O(n^{\frac{1}{2}-\varepsilon})$ for every $\varepsilon>0$. On the positive side, we obtain a linear time algorithm to compute an optimal strong odd coloring for block graphs. From a parameterized perspective, we present an FPT algorithm for STRONG ODD $k$-COLORABILITY when parameterized by treewidth. Moreover, we show that the problem cannot be solved in time $(k-\varepsilon)^{\texttt{tw}}n^{O(1)}$ for every $k\geq3$ and $\varepsilon>0$ when parameterized by treewidth under SETH. Furthermore, we show that STRONG ODD $k$-COLORABILITY does not admit a polynomial kernel when parameterized by feedback vertex set. Lastly, we prove that STRONG ODD $k$-COLORABILITY is W[1]-hard when parameterized by clique-width.

Integer reachability is NP-complete for vector addition systems with states (VASS), but becomes PSPACE-complete in the presence of transfer operations. We refine this complexity gap for single-transfer VASS by identifying structural features of transfers responsible for the increase in complexity. Each system induces a transfer graph whose vertices are counters and whose edges represent possible transfers. We classify its vertices as good or bad, according to the branching and cyclic structure of their reachable subgraphs. Let $b$ be the number of bad vertices. We show that every positive instance admits a polynomially verifiable certificate of size $|I|^{O(b+1)}$, where $|I|$ is the input size. Consequently, integer reachability for single-transfer VASS can be decided in nondeterministic time $|I|^{O(b+1)}$; in particular, it belongs to NP for every class with a bounded number of bad counters. Conversely, we show that bad counters provide sufficient structural power to encode space-bounded computation. For every transfer graph with $b$ bad vertices, we construct a single-transfer VASS that encodes the acceptance of a Turing machine using $b^{O(1)}$ tape cells. This yields PSPACE-hardness for every polynomial-time constructible family of transfer graphs containing linearly many bad vertices. Our results isolate the transfer patterns responsible for the complexity of integer reachability.

Parallel drafting reduces the drafting overhead of speculative decoding for large language models (LLMs), but its gains remain limited by the accepted prefix length. Even when the correct token is present in the candidate pool, a single early selection error prevents subsequent predictions from being used. We propose DRelay, which uses global information from the entire draft block to perform prefix-aware selective repair of candidate selections before target-model verification. DRelay bases its decisions on candidate correlations and the selected path: a global reader extracts predictive information across positions for each candidate. While a causal selector combines candidate-level information extracted by the global read with the tokens selected at preceding positions to determine whether the native choice at the current position is consistent with the global evidence and the selected prefix. It then decides whether to retain or replace the token, thereby repairing early errors and extending the accepted prefix. We further jointly train the draft backbone and the selector, combining candidate-support learning with a repair objective, while weighting the repair loss according to each block position's potential contribution to the consecutive accepted prefix. Across eight diverse benchmarks on an H800 GPU, DRelay consistently improves both average acceptance length and end-to-end decoding performance over DFlash, Domino, and DSpark. Under SGLang serving, DRelay improves average end-to-end speedup over DFlash, Domino, and DSpark by 14.7%-16.8%, 8.7%-9.3%, and 8.1%-9.3%, respectively.

Unmanned aerial vehicle (UAV) photogrammetry is increasingly used in applications requiring high accuracy, such as determining ground surface changes caused by landslides, mining, or microrelief transformation. While acquisition strategies have been widely studied, the influence of the processing workflow-particularly Bundle Block Adjustment parameter settings-remains insufficiently explored. This study addresses this gap through a systematic, full-factorial evaluation of 768 processing variants applied to ten UAV datasets collected over 1.5 years in a 220 ha study area. Eight key parameters were analysed. The results show substantial variability in final 3D accuracy: the best performing variant achieved a root mean square error (RMSE) of 16 mm, whereas the weakest reached 303 mm. The most influential factors were the number of ground control points, the application of additional camera calibration corrections, and the use of the Post-Processing Kinematic GNSS method for determining camera projection center coordinates. The study also evaluates how workflow optimization affects the accuracy of displacement, tilt changes, and horizontal strain determination. While random displacement errors remained stable (RMSE of ~6-7 mm), systematic errors were significantly reduced by over half in all axes, with vertical median absolute error decreasing from 14 mm to 7 mm in the optimized configuration compared to the baseline previously used by the authors. This study provides the first large-scale, practice-oriented assessment of how processing parameter selection shapes the accuracy of both photogrammetric products and deformation indices determination. The results offer actionable guidance for developing more robust and repeatable UAV photogrammetry workflows tailored to high-precision monitoring.

Wireless digital twins (WDTs) are promising enablers for developing and evaluating AI-native radio access networks, yet constructing a high-fidelity WDT typically requires substantial manual effort to integrate heterogeneous information and infer unknown propagation-related parameters. This paper proposes Agentic WDT (AWDT), an end-to-end agentic framework for autonomous WDT construction and calibration using readily available environmental information and measurements readily obtainable from commercial smartphones. AWDT comprises three agents: EnvAgent constructs the propagation environment, OpAgent infers BS and sector configurations, and MatAgent calibrates radio material properties. The agents iteratively refine the WDT using discrepancies between LTE/NR reference signal received power (RSRP) measurements and ray-tracing predictions. Real-world experiments show that AWDT reduces the RSRP prediction MAE from 11.46 to 5.25 dB, demonstrating a substantial improvement in ray-tracing fidelity. Evaluation with an independent measurement system further demonstrates cross-device transferability with lightweight device-specific bias adaptation, highlighting the potential of agentic AI for scalable WDT construction and calibration with limited prior knowledge.

Artificial intelligence systems are increasingly deployed in high impact and safety critical settings, yet security assessment remains difficult to reproduce and defend under audit. Existing approaches often rely on narrative checklists or assessor driven scoring, and they lack an explicit, machine evaluable mapping from observable engineering artefacts to stable technique level outcomes. We present an evidence driven AI security assessment framework that operationalises assessment as a deterministic decision function. The framework normalises heterogeneous artefacts into a project independent Control ID taxonomy scored on a bounded four level ordinal scale, compiles technique level predicates from a pinned MITRE ATLAS snapshot via an explicit mitigation to control mapping, and outputs technique indexed feasibility and impact levels with traceable links back to the triggering evidence. We package all normative choices as a versioned assessment policy object to support repeatable reassessment across snapshots. To ensure semantic correctness, we formally verify boundedness, totality, ordered semantic consistency, and monotonicity of the compiled evaluator over the full declared score domain. We evaluate the framework on five public open source AI projects pinned to explicit repository snapshots, quantify before and after changes under a unified hardening intervention, and validate responsiveness to real engineering changes through fork based implementations of Software Bill of Materials (SBOM) generation and Continuous integration (CI) security scanning gates. Results show consistent downward shifts in feasibility profiles under strengthened observable controls, while worst case residual feasibility persists when technique specific core controls remain absent from the evidence scope.

Tabular foundation models (TFMs) achieve strong predictive performance through in-context learning, yet repeatedly conditioning on labeled data makes inference expensive. Knowledge distillation can reduce this cost by transferring their predictive ability to lightweight, dataset-specific students. However, the dependence of TFM predictions on both a labeled context and a query introduces two design questions: how to construct teacher supervision and whether expanding query coverage improves distillation. We examine these questions across two TFMs and both neural and tree-based students, and derive an effective distillation recipe. The recipe uses the full labeled training set as teacher context and trains students solely on teacher predictions for observed and synthetic queries. On TabArena, the resulting students outperform their supervised trained tuned-and-ensembled counterparts by 57-98 Elo points. Applied unchanged to TALENT, the same recipe improves matched default students on 236-258 of 300 datasets and reduces median primary error by 4.0-6.4%. The distilled students also achieve median inference speedups of 3.0-21.6 times over their teachers, offering a practical trade-off between predictive performance and repeated inference cost. Code is available at https://github.com/nums-ai/TFM_Distillation .

Multimodal large language models (MLLMs) incur substantial inference costs when processing long visual-textual sequences. While existing operation compression methods exploit modality-level redundancy, they largely treat computation within attention heads and shared feed-forward network (FFN) channels as unified units, leaving finer-grained redundancy underexplored. We find that redundancy varies both across modality-interaction paths within the same attention head and across visual and textual executions of the same FFN channel. Based on these findings, we propose Modality-aware Width-wise Operation Pruning (MWOP), which independently prunes visual-to-visual (V2V), text-to-visual (T2V), and text-to-text (T2T) attention paths within each layer, and separately selects FFN channels for visual and textual inputs. A first-order Taylor criterion guides the pruning process, with FFN importance re-evaluated after attention pruning and LoRA-based recovery training. To translate the resulting fine-grained sparsity into practical acceleration, we further develop path-sparse Triton attention kernels and compact visual-side FFN execution. MWOP preserves the token sequence while reducing attention and FFN computation, making it complementary to token compression and enabling simultaneous reduction of sequence length and per-token computation. On LLaVA-OneVision-7B, MWOP alone achieves a $1.6\times$ prefill speedup with 99.7% average performance retention across 12 benchmarks. Combined with two representative token compression methods, it further increases their prefill speedups from $2.0\times$ and $1.9\times$ to $2.9\times$ and $2.7\times$, respectively. Results on Qwen2.5-VL-7B further demonstrate its applicability across architectures. The code is available at https://github.com/EIT-NLP/MWOP.

Weak modal logics can be obtained by restricting the axioms and inference rules governing the modal operator. At the extreme, one may consider the least modal logic, in which no specifically modal axiom or inference rule is imposed. We introduce closure semantics, a new possible-world semantics for such weak modal logics, and establish soundness and strong completeness for a range of systems. We also give truth-preserving translations between closure models and other possible-world semantics, including neighborhood, FMT, and Kripke semantics.

Simulating microstructure evolution requires quantum-mechanical accuracy and mesoscopic reach in length and time scales, a combination that no current method achieves. Classical phase-field models provide this reach, but their accuracy is limited by phenomenological free energies and mobilities. Here we develop a framework for learning ab initio phase-field models, where the mesoscopic equation is not postulated but derived from a Mori-Zwanzig projection of molecular dynamics onto species-density fields under explicit assumptions. The nonlocal free energy and mobility left unspecified by this equation are parametrized by neural networks and learned from short molecular dynamics trajectories generated with machine-learning interatomic potentials of ab initio accuracy. We demonstrate the framework on an iron-boron melt and on hydrogen-helium mixtures under planetary conditions. For iron-boron, the model shows that the melt at the FeB$_4$ composition is spinodally unstable at ambient pressure but stabilized at 10 GPa, offering a thermodynamic rationale for why FeB$_4$ has been synthesized only under high pressure. For hydrogen-helium, the model predicts the immiscibility boundary and captures droplet nucleation and growth in helium-rain simulations of a column corresponding to 2.2 million atoms, far beyond the scale of atomistic modeling at comparable accuracy. Trained across compositions and conditions, such models could provide a mesoscopic counterpart to ab initio molecular dynamics.

Multiple-choice questions are a staple of educational assessment due to their efficiency, but their diagnostic ability is inherently limited. Traditional psychometric techniques infer misconceptions by fitting models to predetermined data patterns, making them ineffective for analysing small, novel, or non-standardised samples where such prior trends are absent. To address this, we introduce a new distractor analysis method based on Information Theory, which treats the randomness in incorrect answers not as noise but as a measurable signal in student answer patterns. By examining the entropy of distractor choices alongside overall accuracy, our approach offers a sample-sized-independent framework for classifying performance. Cross-referencing low-scoring item-level response data against independently documented misconceptions on a standard mechanics inventory shows that, for items identified as most concentrated (lowest degree of randomness), a majority of incorrect responses select the exact response coded to that item's named, interview-validated misconception. Furthermore, items independently flagged elsewhere as diagnostically unreliable are exactly the items the measure identifies as having the highest degree of randomness. The pedagogically actionable cases are those with low degree of randomness in a class's incorrect answers, signalling a shared, re-teachable misconception, whereas a high degree of randomness chiefly serves as the null case against which that concentration is judged. Our analysis is intended as a fast, single-administration triage step: a practical, theory-driven complement to established distractor-analysis methods that helps instructors flag which items warrant closer, more resource-intensive investigation.

Structure formation has been used to test the viability of the IR cutoff in the Tsallis holographic dark energy (THDE) model, and the future event horizon was found to be a viable IR cutoff (Phys. Rev. D 113, 123531 (2026)). Motivated by this, we constrain the HDE, THDE, and BHDE models using the combined Pantheon+ SN Ia, OHD, DESI DR2, and CMB distance prior datasets, and propagate the MCMC posterior samples into the linear growth sector to obtain the predicted $f\sigma_8(z)$ and $S_8(z)$ histories under the smooth dark energy approximation. We find that the growth predictions of HDE and BHDE are nearly indistinguishable, while those of THDE lie lowest. A posterior predictive $\chi^2$ comparison shows that current $f\sigma_8$ measurements are statistically consistent with all three models, whereas the $S_8$ comparison yields markedly larger $\chi^2$ values for HDE and BHDE and a value for THDE close to that of $\Lambda$CDM. As an illustrative diagnostic, we construct the $(f\sigma_8, S_8)$ plane and find that the trajectories of THDE and $\Lambda$CDM reach the vicinity of the current data region, while those of HDE and BHDE do not. Our results indicate that structure growth observables provide a promising probe for discriminating among holographic dark energy models, while a definitive discrimination will require a joint background and growth likelihood and consistent growth measurements.