Papers

Studies of data scaling in electroencephalography (EEG) and related noninvasive recordings suggest that simple wearable devices could eventually reveal more about cognitive states. However, publicly available EEG data are difficult to combine. Different laboratories record different tasks, participants, and devices. Task-agnostic self-supervised learning can exploit such data, but representations learned only with EEG signals have not consistently produced feature representations that are directly effective for decoding the cognitive state. We therefore aimed to incorporate task information by embedding experimental instructions, stimuli and response demands in a common semantic space, and trained an EEG encoder to align signals from a single anterior–posterior differential channel with those task vectors. This framework supports zero-shot prediction without target-condition calibration and few-shot prediction using labeled samples. On a reserved ten-class evaluation, the highest observed zero-shot balanced accuracy was 30.13% (chance 10%). In a separate developmental few-shot experiment, balanced accuracy reached 34.47% using 16 labeled EEG examples per class from another participant, and 74.99% using examples from the test participant's own task recording. Our experiments suggest that training models on datasets with multiple tasks per participant appeared useful, whereas adding datasets containing only one or a few tasks sometimes reduced established performance despite broader task coverage. These exploratory results suggest that EEG datasets that repeat varied cognitive tasks and everyday activities within the same individuals across days, devices and independent studies would be particularly useful for training an EEG foundation model for predicting cognitive states.

Extracting structured information from regulatory documents is a critical but challenging task. Relevant facts are scattered across long, legally dense texts, and traceability of extracted facts to their source is essential for verification. With increasing awareness of the environmental and health impacts of per- and polyfluoroalkyl substances (PFAS), regulatory documents have expanded significantly, varying in scope and product relevance, making interpretation challenging for manufacturers and policymakers. We introduce RegExtract, a benchmark for structured information extraction from PFAS regulatory documents, comprising 488 human-annotated field instances across 20 US state documents and 46 regulations. Each field instance includes a value and a provenance entry, yielding 976 entries in total. We define a typed, multi-field schema targeting key elements such as material definitions, permitted uses, and product exemptions, and propose provenance-aware evaluation metrics tailored to regulatory text. Through experiments with open- and closed-source LLMs across prompting strategies, document condensing methods, and schema partitioning approaches, we find that document identification fields remain challenging, and that field-specific few-shot examples improve extraction. This work lays the groundwork for AI tools for hazardous substance identification, question answering, and compliance monitoring. The code and data are available at https://github.com/IBM/PFAS-regulations.

A central problem in zero-sum theory is to determine the Davenport constant of a finite abelian group. For a finite abelian group $G\cong C_{n_1}\oplus\cdots\oplus C_{n_r}$, with $1<n_1\mid\cdots\mid n_r$, let $\mathsf D(G)$ be the least integer $t$ such that every sequence of $t$ elements of $G$ contains a nonempty zero-sum subsequence, and denote $\mathsf D^*(G):= 1+\sum_{i=1}^r(n_i-1)$. The standard construction shows that $\mathsf D(G)\ge\mathsf D^*(G)$. Equality holds for several important classes of groups but fails in general, and even the question whether $\mathsf D(G)/\mathsf D^*(G)$ is bounded by an absolute constant remains open. In this paper, we prove the upper bound \[ \mathsf D(G)<4.86\,\exp(G)\frac{\log|G|}{\log\exp(G)} \] for every finite abelian group. When $|G|$ is large compared with $\exp(G)$, it improves the previous classical upper bound by a factor of order $\log\exp(G)$. We also prove that, for almost all finite abelian groups (counting isomorphism classes by order), \[ \mathsf D(G)\sim\exp(G)\frac{\log|G|}{\log\exp(G)}\sim\mathsf D^*(G). \] We also study the difference $\mathsf D(G)-\mathsf D^*(G)$: for every function $L$ with $L(x)\to\infty$, almost all groups satisfy $\mathsf D(G)-\mathsf D^*(G)\le L(\mathsf D^*(G))$. However, for every fixed $C\ge0$, the groups with $\mathsf D(G)-\mathsf D^*(G)>C$ have positive lower density.

Weighted sums of random kitchen sinks (W-RKS) is a commonly used feature extraction technique for supervised learning tasks. A major drawback of this method is the redundancy and ineffectiveness of many extracted random features. This paper investigates efficient methods for reducing such redundancy in multi-class classification problems. We specifically formulate feature selection as a variant of the sequentially nested model selection problem and introduce the Successive Empirical Excess Risk Difference Maximization (SEED-MAX) algorithm. It recursively computes empirical risks to identify the most parsimonious yet informative set of features. We derive an analytical lower bound for the probability of correct feature selection by the SEED-MAX algorithm. This lower bound on the probability of correct feature selection can be made arbitrarily close to one by increasing the training dataset size. To further enhance the rate of pruning, we evaluate the features retained by SEED-MAX using the Fisher test (F-Test). Empirical evaluations on UCR time-series classification datasets show that the sequential application of SEED-MAX with the Fisher test achieves approximately 92 percent rate of pruning with negligible degradation in classification accuracy. Implementations of the SEED-MAX and SEED-MAX-F-Test methods are available at https://github.com/OmidiAmin/seed-max-ftest.

We calculate higher-order perturbative corrections for $t{\bar t}H$ production. By adding second-order and third-order soft-gluon corrections to NLO QCD and NLO electroweak (EW) results, we present total and differential cross sections at LHC energies through approximate N$^3$LO (aN$^3$LO) in QCD and NLO in EW. Our aN$^3$LO QCD + NLO EW results provide significant enhancements and a reduced scale dependence for the cross section and the top-quark transverse-momentum and rapidity distributions.

I introduce a new class of cell complexes, called CL-complexes, that provide alternative topological realizations of finite posets. Each realization of a given poset has the same homotopy type as the order complex of that, so in some sense these spaces do not provide anything really new. Their value is that they enable short transparent proofs of standard results in poset homotopy. These include Quillen's Theorem A for finite posets, the Nerve Theorem for good covers of a topological space, and Sheehy's multicover generalization of the Nerve Theorem. Order complexes retain a special status: the order complex of a finite poset admits CL-complex structures both for the poset itself and for its opposite poset. The definition of CL-complexes is motivated by an elementary generalization of the Carrier Lemma of J. W. Walker, which in its original form constructs a canonical homotopy class of maps from a simplicial complex to a space covered by a suitable family of contractible sets. In the generalized lemma, the role played by simplicial complexes and the role played by spaces covered by contractible sets can both be fulfilled by CL-complexes. This sets up an equivalence of categories that tacitly underlies our approach.

Text-to-video (T2V) models are highly sensitive to prompt specificity, while iterative prompt optimization typically improves generation quality only through repeated video generation and vision-language model evaluation at inference time. We propose TIDE, a framework that distills this costly search process into an interpretable and reusable prompt-rewriting Skill. TIDE formulates prompt rewriting as a set of source-conditioned rule modules, each targeting a distinct semantic dimension. Once distilled, the frozen Skill activates the relevant rules and generates both positive and negative prompts in a single LLM call, followed by only one T2V generation, without test-time visual search or backbone fine-tuning. Evolved using only 100 prompts over 14 hours, TIDE achieves prompt-optimization performance comparable to computationally expensive search-based methods. Moreover, when transferred to previously unseen video-generation models, TIDE outperforms existing approaches specifically designed or trained for those models. Code is available at https://github.com/xiningin/TIDE_T2V.

Diffusion language models (DLMs) offer new opportunities for controllable text generation. However, their controllability, defined as the ability to satisfy task constraints while preserving their original capabilities, remains underexplored. DLMs' bidirectional attention allows control tokens to guide generation from arbitrary positions in the output sequence, but effective guidance depends on the compatibility between their content and placement. To this end, we propose AnchorDLM, a framework in which an anchor generator produces an anchor consisting of a short output fragment and its position for each query. The anchor is placed as control tokens at the specified position to initialize the output sequence of a frozen DLM and guide subsequent denoising. We train the anchor generator with reinforcement learning (RL) to jointly optimize anchor content and placement, using rewards computed from the DLM's generated outputs based on answer correctness and constraint satisfaction. Moreover, through the lens of discrete flow matching (DFM), we interpret anchor placement as helping to reshape the initial velocity field that guides generation. Experiments on code generation and mathematical reasoning across five DLM backbones show that AnchorDLM improves controllability and further yields gains in reasoning performance across diverse benchmarks.

We present BSMScanner, a framework in which a beyond-the-Standard-Model (BSM) parameter scan is specified as a set of declarative YAML documents rather than written as a program. The user declares the free parameters and their priors, the mass matrices and the diagonalisations that act on them, the observables, the theory-validity conditions and the likelihood terms. The framework validates this specification, resolves it into a dependency graph and compiles it once into a typed C++ evaluation kernel, which is then driven by an independently chosen scanning engine: uniform sampling, two differential-evolution engines, or a staged basin-discovery orchestrator, optionally followed by Markov-chain posterior refinement. Flavour physics matrix diagonalisation, PMNS and CKM construction in fixed conventions, tabulated oscillation likelihoods and a library of common flavour observables are provided by the framework and imported by name, so that a model file contains only model-specific definitions; quantities without a closed form, such as a relic density, enter through a C++ or Python plugin interface. We describe the specification schema, the observable and likelihood library, the scanning engines and their configuration, and demonstrate the framework on six benchmark models with $4$ to $26$ free parameters, giving for each the specification, the scan configuration, the resulting best fit and analytic cross-checks of the result.

LLM multi-agent systems split work across agents that communicate through text. An agent summarizes its findings, another agent receives the text. This constrains communication to discrete tokens and costs redundant compute, limiting bandwidth and capabilities of multi-agent systems. Recent works demonstrate KV-cache transfer between LLMs but they either require the LLMs to be within the same family or cannot function across context, thus requiring the receiver to prefill text. We present SwaT (Swarm Telepathy), the first cross-family KV-cache transfer method that allows prefill-free communication in multi-family multi-agent systems. In an agent-to-agent communication setting, combining SwaT with recent advances in KV-cache compaction beats text-to-text communication in both accuracy and speed, with a mean of $40\%$ higher accuracy and a mean of $4\times$ faster wall-clock time. Experiments show SwaT achieves high performance on 17 benchmarks across 12 LLMs and 6 families. Additionally, given a sender Vision-Language Model (VLM), SwaT is able to transfer the vision capabilities to a text-only receiver LLM exceeding text-to-text communication despite SwaT being trained on text-only data. SwaT keeps both sender and receiver models frozen without fine-tuning and only requires $<50M$ tokens and $<1$ GPU-day to train.

In this paper, we study finite-velocity random motions subject to a killing mechanism acting at direction changes: at each change of direction, the motion is killed with a fixed probability and remains at its current position thereafter. We first consider the one-dimensional case of a standard telegraph process. We derive the exact distribution of the position at the killing time and show that it is Laplace distributed. We also characterize the law of the process conditional on survival up to a fixed time, proving that it coincides with that of a standard telegraph process driven by a thinned Poisson process with reduced switching intensity. We further derive the full time-dependent distribution of the killed process, including both its absolutely continuous and singular components. We then extend the analysis to a planar finite-velocity random motion with infinitely many possible directions. In this setting, we derive the distribution of the killed process and obtain an explicit representation of its density in terms of modified Bessel functions and a series involving Bessel polynomials.

AI coding agents operate in iterative loops: they modify code, run tools, inspect the results, and decide what to do next. We call the design of this process loop engineering, which includes context management, model selection, compaction, and termination. Existing efficiency work mainly compresses tool observations or routes selected turns to cheaper models, but the practical opportunity for these optimizations remains unclear. We analyze 600 publicly released coding-agent trajectories spanning two agent scaffolds and two language models. Median observations contain only 118-286 tokens, while the large build logs commonly used to motivate compression occur near the ninety-ninth percentile. Moreover, 98.6% of input tokens are served from a prefix cache, so flat input pricing overestimates loop cost by roughly 7x on average under near-ideal caching. We derive and empirically validate a closed-form break-even condition for model delegation. Delegation increases cost in half of the evaluated settings, near the predicted boundary. More broadly, turn count explains 84% of the variance in log trajectory cost, compared with 4% for mean observation size; the most expensive trajectories have the most turns despite having the smallest observations. These associations suggest that observation-focused methods target a secondary cost factor, and identify unnecessary turns as the larger and largely unexamined opportunity, though we do not establish that turns can be removed while holding task success fixed. We release TurnCost, a tool for re-scoring trajectory corpora under alternative caching, pricing, compression, and routing policies.

Quantization plays a crucial role in efficiently deploying deep learning models on resource-constrained devices. Post-training quantization does not require access to the original dataset or retraining the full model. Current methods that achieve high performance close to baseline results typically require INT8 fixed-point integers. However, achieving higher model compression with lower bit-widths often leads to significant performance degradation. In this paper, we propose VLCQ, a post-training quantization method that relaxes the fixed-point encoding constraint, enabling more flexible exploration of quantization techniques through variable-length encoding. This approach achieves competitive or improved performance while reducing bit-widths, without requiring access to training data or model fine-tuning. Extensive experiments on image classification, segmentation, and object detection demonstrate that VLCQ improves model compression and accuracy compared with state-of-the-art post-training quantization approaches. For per-channel quantization, VLCQ surpasses FP32 accuracy and Piece-Wise Linear Quantization (PWLQ) on most evaluated models, achieving up to 6x compression compared with FP32 and up to 1.7x compared with PWLQ. When prioritizing compression with limited accuracy loss, VLCQ achieves up to 12.25x compression compared with FP32 with less than 4% performance degradation. For per-tensor quantization, VLCQ is competitive with Data-Free Quantization (DFQ), while providing greater flexibility to achieve lower bit rates across different tasks and models.

Argument structures represent how individual arguments interact within a discourse. However, leveraging these structures to infer meaningful knowledge remains challenging, as it requires a formal representation of how argumentative influence propagates across the structure. We propose Neuro-symbolic Argumentation Reasoning System (NARS) for performing persuasiveness detection directly from online discussions. The neural component employs retrieval-augmented iterative LLM reasoning to extract bipolar argument structures, while the symbolic component applies gradual argumentation semantics to propagate argumentative influence over the structure. Using the Webis-CMV-20 corpus, we construct CMV2BAS, a dataset of long, multi-threaded ChangeMyView discussions annotated with LLM-derived bipolar argument structures and labeled with delta events (view change vs.\ no change). We investigate whether gradual argumentation semantics can effectively detect persuasiveness, whether augmenting them with argument quality and graph-based centrality improves performance, and whether symbolic reasoning remains necessary in the presence of modern LLMs. NARS achieves F1=0.57 and AUC=0.56 under its best configuration, outperforming zero-shot Qwen3.6:27B and GPT-oss:20B. We identify support-dominant discussions and long argumentative chains as key failure modes limiting attack influence propagation. Our work demonstrates the feasibility of operationalizing formal gradual argumentation semantics over noisy real-world discourse for scalable macro-level persuasion analysis, while highlighting key challenges and directions for future research in neuro-symbolic argumentation reasoning. We release both CMV2BAS dataset and the NARS framework for public use.

Plasma-neutral interactions critically affect the edge dynamics of magnetically confined fusion devices. Neutral atoms modify ion and electron velocity distributions, potentially affecting local instabilities Standard fluid models simplify these processes (namely charge exchange, ionization and recombination) using Maxwellian assumptions and linearized source terms. We revisit these assumptions for a hydrogen plasma interacting with hydrogen atoms. We derive plasma-neutral interaction terms in the kinetic and gyrokinetic frameworks. Accounting for velocity-dependent differential cross-sections, the model captures the kinetic impact of inelastic collisions on ions and electrons beyond energy and momentum exchange. We demonstrate that the fluid moments of the kinetic model reduce to standard fluid source terms under the appropriate assumptions of Maxwellianity and linearity. We compare the kinetic model with fluid models through their effects on distribution functions and fluid moments. Substantial discrepancies are observed, even for Maxwellian distributions. For charge exchange, the kinetic model yields momentum and energy exchange rates up to twice those predicted by fluid descriptions. We show that a fluid treatment of ionization greatly overestimates the mean kinetic energy of the produced electrons. These results highlight the limitations of fluid descriptions and the importance of kinetic effects in edge-plasma momentum and energy exchange.

Corrosion of structural alloys remains a critical challenge for advanced molten salt reactors, where aggressive fluoride salts promote continuous dissolution at the salt-metal interface. While most mitigation strategies emphasize thermodynamic stability, kinetic control of atomic transport has received comparatively less attention. Here, we propose a corrosion mitigation strategy based on diffusion control and vacancy trapping by dilute yttrium (Y) solutes and yttrium oxide (Y$_2$O$_3$) dispersions in Ni alloys. A combined computational and experimental approach integrates density functional theory (DFT), self-consistent mean-field (SCMF) diffusion modeling, and an impurity trapping model to quantify the impact of defect trapping on effective vacancy mobility. Strong first-nearest-neighbor vacancy-Y binding, off-lattice vacancy configurations, and elevated rotational and dissociation barriers result in pronounced suppression of vacancy diffusivity, while Y$_2$O$_3$-associated sites treated as fixed traps in the model further reduce the mobile vacancy population. Ni-Mo (NM), Y-modified (NMY), and Y$_2$O$_3$-containing (NMYO) alloys were tested in molten fluoride salt at 750{\deg}C. Corrosion penetration decreases systematically from NM to NMY to NMYO, consistent with the predicted reduction in vacancy-mediated transport. The same trend is observed in mass change, with net weight loss decreasing from 0.52% for NM to 0.23% for NMY and 0.09% for NMYO. These results are consistent with an alloy-side transport contribution from defect trapping and reduced atomic transport. This study therefore suggests point defect engineering as a promising alloy design strategy for extreme nuclear environments.

Among connected, planar sets, the ratio between the isoperimetric deficit and the squared barycentric asymmetry is known to be bounded from below by a strictly positive constant (as proved in recent papers by Bianchini-Croce-Henrot and Gambicchia-Pratelli). This opens the natural question whether the infimum is attained at some set. In this paper, we prove that a minimizer for this ratio actually exists, and moreover it is simply connected.

We study generalized verifiable disclosure by a sender with type-independent preferences. We first characterize all perfect Bayesian equilibria as partitions of the sender's types into coalitions ordered by payoff. An equilibrium survives our coalition-proofness criterion if no group of types can profitably coordinate an announced deviation that only they are able and willing to join. Coalition-proof equilibria are greedy partitions: at each step, a coalition of the remaining types attains the highest payoff consistent with equilibrium. In our general setting, coalition-proof equilibria may fail to exist. We establish existence under restrictions on preferences and evidence: quasiconcavity with message completeness, betweenness, cheap talk with message completeness, and payoff degradation. For a disclosure game with maximally rich evidence, we show that the sender's ex-ante payoff in a coalition-proof equilibrium is a piecewise linear tent over the belief simplex, analogous to the concave closure in Bayesian persuasion and the quasiconcave closure in cheap talk.

We study the large-time behavior of the $L^2$-norm of solutions to the free wave equation on $\mathbb{R}^N$ and, more generally, their homogeneous Sobolev norms of negative order. We refine and extend the results of R. Ikehata [J. Hyperbolic Differ. Equ. 20 (2023)] who identified and quantified the infinite-time $L^2$-growth of free waves in dimensions $N=1,2$. For finite-energy data with $u_1\in L^1(\mathbb{R}^N)$, we derive exact first-order asymptotics: when $P=\int_{\mathbb{R}^N}u_1(x)\,dx\neq0$, the squared $L^2$-norm grows linearly for $N=1$ and logarithmically for $N=2$, with a leading coefficient depending only on $P$. In particular, no weighted moment assumption on $u_1$ is required; $u_1\in L^1\cap L^2$ suffices. More generally, we prove that the $L^2$-norm is bounded, and hence convergent, if and only if $u_1\in\dot H^{-1}(\mathbb{R}^N)$. We further establish first-order asymptotic profiles in negative-order homogeneous Sobolev spaces, covering polynomial growth, logarithmic growth, and convergence to a finite limit according to the Sobolev index. These results reveal a unified low-frequency mechanism underlying the low-dimensional $L^2$-growth. We also derive second-order asymptotic expansions in dimension one and at the critical Sobolev index in dimensions $N\ge2$. Finally, in dimension two, the exact $L^2$-asymptotics yield a sharper local energy decay estimate with an explicit upper asymptotic constant independent of the localization radius.

This article is concerned with the problem of nonlinear (stochastic) filter stability for a discrete time hidden Markov model (HMM) with finite state and observation spaces. For this purpose, a comprehensive duality theory is developed whereby the dual control system is a backward stochastic difference equation (BSDE). The duality is expressed in two inter-related forms. In its variational form, the minimum mean squared error of optimal filtering equals the value of a dual optimal control problem. In its structural form, controllability (resp. stabilizability) of the BSDE and observability (resp. detectability) of the HMM are dual properties. The dual optimal control problem supplies the Lyapunov function and the dissipation: the former is the value function (a variance), the latter the optimal running cost accumulated over the horizon. The main result is an invariance principle: under any one of three model properties—ergodicity of the Markov matrix, or observability or detectability of the HMM—zero dissipation over a finite horizon forces the Lyapunov function to vanish. A converse is also obtained: detectability is necessary. Two applications are given. The first is filter stability in chi-squared divergence: the invariance principle is applied in the manner of LaSalle, to limits of trajectories along which the dissipation vanishes asymptotically, the role of the model properties being to render the zero-dissipation set trivial. The second is a quantitative form of the invariance principle, through a Poincare constant, from which convergence rates follow. Both the model properties that govern stability and the Lyapunov function that certifies it are thus obtained from a dual optimal control problem, as in the classical theory for the linear Gaussian model.

Visual agents that interleave reasoning with tool invocation have posed a promising direction for multimodal tasks. However, learning effective tool invocation usually relies on curated question-answer pairs or expert demonstrations at scale, which are expensive to acquire. In this study, we propose Marvo, a simple-yet-effective self-evolution approach that enables learning tool invocation from unlabeled images. Marvo consists of two agents: (1) a challenger that generates question-answer pairs from unlabeled images, and (2) a solver that learns tool invocation by solving questions through reinforcement learning. To encourage effective tool invocation, a good training question should be difficult enough to warrant tool use, yet remain within the solver's capabilities. To this end, we introduce a margin reward to the challenger, defined as solver's performance difference with and without access to tools. With the margin reward, we create an adaptive curriculum that encourages the solver to learn effective tool invocation. Across benchmarks covering agentic reasoning, visual search, mathematical, and general real-world reasoning, Marvo improves average accuracy by up to 7.1% — demonstrating margin reward as a preferred way to scale question complexity without human supervision.

Quantum generative models face growing qubit requirements and concentrated loss landscapes when learning long time series. Here we introduce holographic quantum circuit sampling (HQCS), which combines mid-circuit measurements, qubit reuse and independently parameterized temporal blocks to generate trajectories with a fixed number of qubits. For Haar-random HQCS and a normalized exponential kernel, we prove inverse-polynomial bounds on maximum mean discrepancy loss variance, contrasting with exponential concentration in global Haar-random generators. For generic members of a family of nonstationary Markov processes, temporal parameterization reduces exact-generation gate counts from quadratic to linear in sequence length relative to a specified class of repeated shared-block generators. Simulations demonstrate learning of Gaussian autoregressive processes and historical Volatility Index (VIX) trajectories, with the latter achieving over tenfold lower maximum mean discrepancy than the tested classical benchmarks. These results establish a framework for quantum time-series generation with fixed qubit requirements and provable avoidance of exponential loss concentration as sequence length grows.

As text-to-speech (TTS) systems continue to advance, provenance tracking and security verification of synthesized speech have become increasingly important. Audio watermarking provides active provenance by embedding imperceptible, machine-recoverable information into speech. However, existing post-hoc methods typically modify the waveform after synthesis, leaving the watermark vulnerable to compound perturbations. To address this limitation, we propose GAMW, a watermark injection framework integrated into the TTS generation process. Before waveform synthesis, GAMW embeds a content-adaptive multi-bit watermark residual into the generated representation, and its decoding objective combines focal modulation with dynamic per-bit weighting. We train and evaluate GAMW separately on two TTS backbones. With a 16-bit payload, GAMW improves decoding accuracy over the strongest baseline by 9.2 and 23.5 percentage points under individual and compound perturbations, respectively, while maintaining competitive speech quality. Reliable clean-speech decoding is also retained at higher payloads. Code is available at https://github.com/ZConJugaTe/GAMW.

Adapting to a new world and reasoning correctly about which outcomes are possible under rules one has not met before is a basic capacity underlying general intelligence. We study it in small discrete worlds: sets of grid configurations consistent with a shared set of rules, concretely Sudoku-like puzzles built from a common base and a handful of extra rules. We train a single depth-recurrent model on many combinations of rules and test it at four levels of novelty: a new puzzle, a new combination of known rules, a pairing of rules never seen together, and a rule instance not seen at all during training. The model receives the active rules either as a symbolic description or as two solved examples, and a model that gets no rule information is used as a baseline. At recombining rules it was trained on, the description matches or exceeds examples ($99\%$ against $95\%$ on new combinations). On rule instances whose components never co-occurred in training, the description falls below the no-information baseline while examples stay above it. The framework allows us to test many such generalization regimes in more systematic way than benchmarks such as ARC-AGI and provides new ground for evaluating depth-recurrent reasoning models, which are typically tested on saturated Sudoku benchmarks.

We present a typed proof-term calculus for intuitionistic intensional first-order logic with Hilbert's epsilon-operator and prove its soundness and completeness with respect to hyperdoctrinal models with thin bicartesian closed setoid categories (BiCCCs) as fibres and formula-indexed witness selections. The calculus provides an explicit internal language for these epsilon-hyperdoctrines: for each fixed typed signature, its syntactic term model is initial for syntax-indexed, interpretation-preserving model morphisms, with uniqueness up to setoid equality. The nested epsilon syntax is organized by a well-founded dependency relation on formula labels. Concrete countermodels exhibit failure of excluded middle and show that formula conversion need not identify the selected witnesses. We also compare the full calculus with its epsilon-free and epsilon-only fragments. The principal constructions and metatheorems are formalized in Rocq. With setoid BiCCCs, proof irrelevance is a choice of equality, not the erasure of proofs: the proofs and their categorical operations remain available as the basis for a future proof-relevant categorical semantics.

Squeeze-and-Excitation (SE) modules improve convolutional networks by modeling channel-wise dependencies, but global average pooling collapses spatial structure into a single per-channel descriptor, limiting their effectiveness for dense prediction tasks such as image segmentation. We generalize SE by replacing global pooling with spatially guided pooling, enabling region-aware channel recalibration through four modules with complementary cost profiles. BASE formalizes Tiled Squeeze-and-Excite using a regular grid; SPASE pools over SLIC superpixels for image-adaptive regions; MAPSE obtains regions from semantic maps produced by a frozen, general-purpose network; and PLEASE learns soft region assignments end-to-end through a lightweight assignment network. Each channel receives one gate per region, so attention at each pixel reflects its region membership, bridging channel and spatial attention. PLEASE also produces its soft assignments as a byproduct of the forward pass, providing structured spatial interpretability without post-hoc analysis. The modules offer complementary computational tradeoffs: some add essentially no parameters, while others add no measurable latency. Across binary segmentation on CUB, DUTS, and ECSSD, semantic segmentation on PASCAL VOC 2012 and ADE20K, and classification on mini-ImageNet, our modules outperform SE and related spatial-channel attention variants on most benchmarks, achieving the majority of best results.

Machine-learned interatomic potentials (MLIPs) accelerate molecular dynamics (MD) simulations by learning the energies and forces that govern atomic motion. Recent work has shown that relaxing physical inductive biases can improve computational efficiency and predictive accuracy, raising the question of whether energy-and-force matching alone is sufficient to ensure accurate sampling. We show that, for unconstrained hypothesis classes, the relationship between prediction error and sampling accuracy is exponentially ill-conditioned in dimension. In particular, prediction errors that decay exponentially with dimension can coexist with nearly maximal total variation distance between the learned and target distributions. We identify two failure mechanisms. The first is kinetic trapping, where relaxation to an accurate learned equilibrium requires exponentially long times, even when simulations are initialized from training configurations at which energies and forces are matched exactly. The second is equilibrium mismatch, where the learned stationary distribution itself remains far from the target. Under explicit assumptions, we show that equilibrium mismatch can persist even in local, symmetry-preserving models, where small per-atom errors coexist with a distributional discrepancy that approaches its maximum as system size increases. Finally, we establish global energy-control conditions that guarantee equilibrium accuracy and identify additional conditions required for rapid relaxation.

A brain tumor is an unusual growth of cells in the brain that can be cancerous or not, affecting normal brain function and causing serious risk to health. Existing solutions for brain tumor classification, such as expert manual evaluations of magnetic resonance imaging (MRI) scans, are generally inconsistent as different professionals have different criteria and can be resource-intensive, leading to delayed diagnoses. Additionally, while deep learning models have demonstrated potential, they typically require significant computational resources, motivating the need for more efficient and accessible diagnostic tools that can be deployed in real-time healthcare settings. To address these limitations, this study presents BrainAid, which makes efficient use of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for effective classification of four-class classification of glioma, meningioma, pituitary tumor, and no-tumor MRI scans, outperforming all the standard models such as VGG16, Inception-V3, ResNet-50 and so, while integrating Explainable AI (XAI) to improve interpretability. The BrainAid is trained to classify tumors with a high precision of 98%, achieving an F1 score of 98% and an accuracy of 98% using publicly available dataset. The model is deployed on a Raspberry Pi, demonstrating efficiency in low-resource environments. This system shows the feasibility of real-time, edge-based tumor classification, providing quick diagnostic support for medical professionals and enhancing patient care in resource-limited healthcare facilities and underserved areas.

Plasma experiments observe coupled physical states through sparse, heterogeneous, and incomplete diagnostics. We present a proof of concept toward a multimodal foundation model for plasma state reconstruction during magnetic reconnection, motivated by the Facility for Laboratory Reconnection Experiments (FLARE). Using 250 synthetic VPIC simulations at fixed reference electron beta $\beta_{e0}=0.2$, with 225 training runs and 25 disjoint validation runs, we train a 5.5M-parameter residual 3D U-Net to reconstruct three magnetic-field components and electron density from incomplete observations. The model combines spatial-only pooling, which preserves all 24 temporal frames, with global spatiotemporal self-attention and rotary positional encoding. Independently masking magnetic and density modalities across random, grid, block, and temporal patterns enables sparse sensing, cross-modal completion, density super-resolution, and temporal completion with a single set of weights and no task-specific retraining. With density fully hidden, just 0.1% magnetic spatial visibility reduces median density NRMSE from approximately 0.9 to 0.1. Conversely, observed density improves current-density reconstruction when magnetic data are absent or extremely sparse. With no magnetic observations, roughly 1% density visibility approaches the density reconstruction accuracy achieved with fully observed magnetic fields, although current structure remains more dependent on direct magnetic measurements. Full magnetic conditioning also stabilizes completion of missing future density frames. These results demonstrate bidirectional cross-modal predictive coupling within this simulation regime. Generalization across plasma regimes or to experimental measurements remains untested.

We calculate the masses of the $J^P=\frac{1}{2}^-$ hidden-charm pentaquark states in the $SU(3)$ decuplet within the heavy pentaquark chiral perturbation theory (HPChPT) up to the next-to-leading order (NLO). The one-loop self-energy corrections from the $\pi$, $K$, and $\eta$ mesons are evaluated. Taking $P_\psi^N(4312)$ and $P_{\psi s}^\Lambda(4459)$ as inputs, we predict the masses of the doubly-strange and triply-strange decuplet members $P_{\psi ss}^N$ and $P_{\psi sss}^\Lambda$ to be 4.607 GeV and 4.755 GeV, respectively. The mass splittings between adjacent strangeness layers are nearly constant at approximately 0.148 GeV. This equal-spacing pattern follows directly from the analytical structure of the NLO chiral corrections, while its actual magnitude is set by the low-energy constants (LECs). It indicates that the Gell–Mann–Okubo (GMO) relation, expected at the leading order of SU(3) flavor symmetry breaking, persists to good accuracy at the NLO.