arXiv subjects

Zhen Zhang

Publications and source records attributed to Zhen Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Differentiable Fuzzy Inference Layer: A Monotone, Compositional Ordinal Reasoning Head for Large Language Models

A state-of-the-art language model asked to interpret "most of most students passed" typically answers "most," though composing two instances of "most" yields a proportion closer to "some." We trace this failure to an architectural choice rather than a data deficit: standard classifier heads treat ordinal categories as independent labels, with no mechanism to respect their natural ordering or compose them algebraically. We introduce the Differentiable Fuzzy Inference Layer (DFIL), a dual-path prediction head pairing a standard classifier with a scalar-bottlenecked branch grounded in a bank of ordered membership functions. DFIL supplies two structural primitives that a label-only head cannot inherit: monotonicity in the underlying quantity, and compositional reasoning via t-norm operations without any compositional training data. The scalar branch additionally provides an interpretable interface for analyzing residual errors. We instantiate DFIL on ordinal natural-language tasks across diverse LLM families.

cs.CL↗

arXiv2026

The Cost of Conservation: Coordination-Memory Laws for Exact-Support Generation

Many AI systems make decisions locally, even when every realized output must obey an additive conservation law, such as selecting exactly a fixed number of items. This constraint can be statistically invisible: small subsets of a balanced fixed-budget output look increasingly independent, yet communication-free coordinate-parallel generation needs exponentially many pre-shared plans, while a sequential exact sampler needs only logarithmic memory. We study product measures conditioned on additive conservation laws in the intermediate regime where one plan is selected before a fixed-order pass, every plan is a bounded-state stochastic executor whose support is entirely legal, and the mixture of plan laws approximates the target distribution in total variation. Our main result identifies the optimal asymptotic selector rate, up to constant factors, with the killed spectral profile of the conservation-difference walk. The resulting coordination cost decreases as an inverse power of live-state width, with an exponent determined by intrinsic conservation rank rather than alphabet size; the law extends to noncentral budgets and heterogeneous local scores. The converse is driven by a state-versus-resource-sum obstruction, while a rate-matching construction compiles discrepancy control into exact-support finite-state plans. Complementary results characterize block-parallel plan complexity and the benefit of programmable output order. Together, these results show precisely how online memory substitutes for front-loaded coordination in exact-support generation.

cs.AI↗

arXiv2026

MathGen: Revealing the Illusion of Mathematical Competence through Text-to-Image Generation

Modern generative models have demonstrated the ability to solve challenging mathematical problems. In many real-world settings, however, mathematical solutions must be expressed visually through diagrams, plots, geometric constructions, and structured symbolic layouts, where correctness depends on precise visual composition. This naturally raises the question of whether generative models can still do so when the answer must be rendered visually rather than written in text? To study this problem, we introduce MathGen, a rigorous benchmark of 420 problems spanning seven core domains, including 350 Clean-Scene problems and 70 paired Open-Scene problems. Each problem is evaluated under a Script-as-a-Judge protocol with problem-specific verification criteria implemented through reusable executable scripts for deterministic and reproducible evaluation. Experiments on representative open-source and proprietary text-to-image models show that mathematical fidelity remains a major bottleneck: even the best closed-source model reaches only 53.7% overall accuracy, while open-source models achieve just 1--11%, often near 0% on structured tasks, particularly those requiring precise geometric and functional rendering. Overall, current T2I models remain far from reliable at even elementary mathematical visual generation.

cs.CV↗

arXiv2026

CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts

Contrastive vision-language models, such as CLIP, have garnered considerable attention for various downstream tasks, mainly due to the remarkable ability of the learned features for generalization. However, the features they learned often blend content and style information, which somewhat limits their generalization capabilities under distribution shifts. To address this limitation, we adopt a causal generative perspective for multimodal data and propose contrastive learning with data augmentation to disentangle content features from the original representations. To achieve this, we begin with exploring image augmentation techniques and develop a method to seamlessly integrate them into pre-trained CLIP-like models to extract pure content features. Taking a step further, recognizing the inherent semantic richness and logical structure of text data, we explore the use of text augmentation to isolate latent content from style features. This enables CLIP-like model's encoders to concentrate on latent content information, refining the learned representations by pre-trained CLIP-like models. Our extensive experiments across diverse datasets demonstrate significant improvements in zero-shot and few-shot classification tasks, alongside enhanced robustness to various perturbations. These results underscore the effectiveness of our proposed methods in refining vision-language representations and advancing the state-of-the-art in multimodal learning.

cs.CV↗

arXiv2026

Probing Variations in Earth's Ionosphere Using Pulsars

We present high cadence 10-minute rotation measure (RM) monitoring of PSR~J0814+7429 using the LOw-Frequency ARray, aiming to probe ionospheric variability along the pulsar line of sight (LoS). By separating the ionospheric contribution from the observed RM, we quantitatively reconstruct the diurnal variation of the ionospheric electron density along the pulsar LoS based on the World Magnetic Model. The derived variations exhibit clear solar-driven modulation, including the ionospheric noontime bite-out phenomenon, and show good agreement with the LoS total electron content reconstructed from independent global vertical total electron content maps. These results demonstrate the feasibility of using pulsars as probes of temporal variations in the electron density of the Earth's ionosphere.

astro-ph.HE↗

arXiv2026

Hybrid Probabilistic Zonotopes for Identifiable and Refinable Predictive Uncertainty

Probabilistic prediction heads in neural networks typically output either a Gaussian mixture or a single conformal region. Neither separates the distinct sources of uncertainty often present in real prediction tasks: a discrete choice among modes, bounded systematic drift within the chosen mode, and irreducible stochastic noise. We introduce the Hybrid Probabilistic Zonotope (HProbZ), an output head that represents these three sources as binary, bounded, and stochastic generators of a zonotope, and admits a closed-form likelihood by convolution. Sharing the bounded generator across prediction steps couples future predictions algebraically, so observing one step refines the predictive distribution at every remaining step in a single forward pass. We establish that the three generators are identifiable from the likelihood up to permutation, and that an HProbZ density is representationally distinct from any finite Gaussian mixture. The same shared structure provides analytic per-mode risk and distribution-free multi-modal conformal sets at inference time. Empirical analysis on representative prediction benchmarks supports the effectiveness of the design relative to same-encoder mixture baselines, while offering structural properties that mixture or convex-conformal predictors do not jointly provide.

cs.LG↗

arXiv2026

Matrix Zonotopic Attention: A Context-Adaptive Value Projection for Set Transformers

Multi-head attention combines an input-dependent softmax routing with an input-independent linear value projection, so the per-sample operator mapping aggregated values to outputs is the same for every input set. We study the consequences of this asymmetry for permutation-invariant set targets. We introduce the Transformation Degrees of Freedom (TDOF) of a target operator, a complexity measure counting the input-dependent directions an exact representation requires, and present a depth-separation analysis showing that context-rigid attention needs depth proportional to the target's TDOF, whereas a single layer with a context-adaptive value family can represent the same target. Building on this analysis, we propose Matrix Zonotopic Attention (MZAttn), which replaces the fixed value projection with a context-adaptive matrix-zonotope family: a centre matrix plus a sum of generator matrices weighted by input-dependent gates. The construction reduces to standard multi-head attention at initialisation, preserves permutation equivariance, and admits a data-driven reachability interpretation. Experiments on a range of set-prediction tasks are consistent with the TDOF prediction that the architectural advantage is selective: it appears on targets that depend on the input set in a high-rank, sparsely combinatorial way, and is small on aggregate-statistic targets where parameter-matched standard attention is already competitive.

cs.LG↗

arXiv2026

Double Parton Scattering Effects on the Measurement of the W-Boson Mass

Double parton scattering (DPS) corresponds to events where two parton-parton scatterings occur in a single hadron-hadron collision. The DPS effects may arise from the spectator scatterings that are somewhat related to semi-hard QCD activities. In this work, we investigate the DPS effects on the W-boson mass measurements. Especially, our analysis reveals that the DPS effects contribute additional missing transverse momenta from spectator scatterings as well as relevant inclusive cross sections, potentially altering the distribution of total missing transverse momenta. Consequently, the DPS effects have the potential to cause an increase in the measured W-boson mass by the CDF detector, which helps to understand the deviation of the CDF-II measurements from other measurements and the predicted value in the Standard Model. This could be further validated in upcoming studies, such as the W-like analyses of the Z boson production, which will potentially open new avenues for probing QCD dynamics in the semi-hard regime.

hep-ph↗

arXiv2026

Optimizing Parameterized Physics-Informed Neural Networks to Solve Multilayered Static Linear Elastic PDEs

Designing multilayered materials for controlled deformation is heavily bottlenecked by the immense computational costs and time of traditional, industrial-grade finite element methods (FEM). This excessive expense severely limits design space exploration and gradient-based optimization. To streamline the workflow, we propose a framework for parameterized physics-informed neural networks (P2INNs). The framework encompasses three core components: I. a FEM baseline with a Hex8 element and trilinear basis functions, II. a P2INN displacement field model across variable material stiffness and layer thicknesses, and III. FEM-referenced evaluation. The PINN enforces static linear elasticity via Navier-Cauchy residuals with training strategies that prioritize physics, including layerwise PDE decomposition, interface continuity penalty, and compliance aware scaling. For one-layer, pure physics-driven configurations, the PINN achieves a mean volume MAE of 1.56% and worst-case volume MAE of 2.87% against the FEM benchmark. For three-layer configurations with controlled supervised training, the PINN achieves a mean volume MAE of 2.53% and worst-case volume MAE of 4.66%, meeting the near-5% worst-case volume-MAE target for preliminary design-space exploration. The trained P2INN model remains a lightweight model with a simple forward pass for future calculations, creating an alternative to traditional FEM. By drastically reducing the need for expensive FEM evaluations, this approach could accelerate the inverse design and optimization of advanced layered architectures for protective structures in various load-heavy or potentially collision-heavy fields.

cs.CE↗

arXiv2026

Effects of light-cluster degrees of freedom on collective flows in heavy-ion collisions at FOPI energies

Within a lattice Boltzmann-Uehling-Uhlenbeck transport model coupled to a kinetic approach for light-cluster formation, we investigate the impact of explicit light-cluster degrees of freedom on collective flows in Au+Au collisions at FOPI energies with beam energies $E_{\rm beam}$= $120$--$1500 A$ MeV by using a density-, momentum-, and isospin-dependent N$5$LO Skyrme pseudopotential. We first benchmark the kinetic approach by comparing the calculated light-cluster yields with FOPI data in central Au+Au collisions. We then analyze the collective flows of protons and light nuclei (deuterons, tritons, $^{3}\mathrm{He}$, and $^{4}\mathrm{He}$) in mid-central collisions. For protons, calculations with and without dynamical light-cluster degrees of freedom are compared to quantify the influence of dynamical cluster formation on proton directed ($v_1$), elliptic ($v_2$), triangular ($v_3$), and quadrangular ($v_4$) flows. We find that the dynamical light-cluster effect appreciably modifies proton $v_1$--$v_4$ flows at $E_{\rm beam}=120$--$150 A$ MeV, remains visible at $E_{\rm beam}=250$--$400 A$ MeV, and gradually weakens at $E_{\rm beam}\gtrsim 600 A$ MeV. For light nuclei, the kinetic approach captures the overall beam-energy dependence of the FOPI flow data, with better agreement for $E_{\rm beam}\geq 400 A$ MeV. We further examine the nucleon-number scaling of $v_2/A$ in both model calculations and experimental data, finding that the kinetic light-cluster formation approach qualitatively reproduces the observed scaling behavior. These results highlight the importance of a dynamical treatment of light-cluster formation for interpreting collective flows in heavy-ion collisions below about $600 A$ MeV, although the clustering effects on proton flows are minor at higher collision energies.

nucl-th↗

arXiv2026

Certifying Plans under Model Mismatch: A Trilemma for Reachability from Scarce Data

Sim-to-real policies are designed under nominal dynamics, but target-system trials may yield only a few isolated one-step transitions. We study pre-execution certification of a fixed control sequence, such as an action chunk produced by a learned policy. If the sequence reaches an unobserved state-input region, the observations remain consistent with target systems whose trajectories separate along it by an arbitrarily large amount. Any deterministic certifier sound for all of them must then decline to certify or return a reachable tube with arbitrarily large projected width. For bounded smooth classes of the target-nominal model error, we derive a finite plan-dependent projected-width lower bound. These results expose a trilemma among uniform trajectory containment, finite projected width, and unrestricted model-error behavior beyond the observations. ForeReach requires a supplied componentwise Lipschitz bound on the model error. Observed transition pairs can refute this declaration but cannot establish it outside the observed locations. Conditional on a valid declaration, our method constructs a set-membership envelope for the model error, propagates a zonotopic reachable tube, and certifies only when propagation remains within the certification domain and every projected tube slice avoids the unsafe set. In two benchmark systems, calibration baselines may remain narrow after losing trajectory containment outside data support, whereas our method declines to certify unsupported sequences and recovers certification when relevant target data and sufficient obstacle clearance are available.

cs.RO↗

arXiv2026

Probabilistic Reachable-Action Verification of Visuomotor Policies via Set-Based Training

Reachability analysis for visuomotor policies is difficult because large visual encoders make end-to-end set propagation computationally expensive and excessively conservative. We therefore freeze the visual encoder and confine set propagation to a low-dimensional interface between it and the downstream policy, with the interface set calibrated from held-out camera-pose perturbations. Propagating this set through the policy with zonotopes yields a terminal output-enclosure width that set-based training optimizes directly. During evaluation, camera-pose perturbations are sampled from the prescribed distribution, and rollout-level split conformal calibration converts the resulting action-deviation scores into a probabilistic reachable-action radius with finite-sample coverage. In controlled manipulation experiments, set-based training reduces this radius while preserving closed-loop task capability, and matched behavior-only, observational-consistency, and pointwise-adversarial controls all leave a larger radius.

cs.RO↗

arXiv2026

A lab-on-a-silicon-chip platform for all-electrical antibiotic susceptibility tests with a sample-to-results time within 20 minutes

Rapid antibiotic susceptibility tests (ASTs) are essential for quick selection of effective drugs to treat bacterial infections at an early stage. However, the most widely used phenotypic ASTs in clinical practice often require 24 - 48 hours of pre-culture enrichment and 4 - 20 hours of testing. They are too slow to guide therapy, even with the most rapid protocol. Here, we report a lab-on-a-silicon chip (LOSC) system, which integrates arrays of silicon nanowire field-effect transistor (SiNWFET) sensors with high-throughput cell-collection microfluidics for rapid ASTs. The microfluidics concentrate bacteria into picoliter-scale chambers within minutes, eliminating the need for pre-cultivation. Embedded SiNWFETs sensitively track antibiotic-induced metabolic pH shifts, enabling AST by detecting rapid metabolic inhibition in susceptible bacteria. Using an unbuffered culturing medium, LOSC achieves sample-to-result times within 20 minutes for clinically isolated E. coli strains. With its electrical readout and compact design, LOSC offers a low-cost, rapid, and portable AST solution for point-of-care diagnostics.

physics.ins-det↗

arXiv2026

Quarkyonic Stars with Strangeness

We propose an extension of the quarkyonic matter framework that includes $u$, $d$, and $s$ quarks and the full baryon octet. Within this extended framework, we impose beta-equilibrium between baryons and leptons, while determining the quark fractions from the constituent quark contents of baryons. The hadronic sector of octet baryons is described by a recently developed density, momentum and isospin dependent effective interaction based on the N3LO Skyrme pseudopotential, whereas quarks and leptons are treated as free particles. We find that the quarkyonic mechanism can obviously reduce the critical density for hyperon appearance in neutron stars due to the fact that the nucleons are displaced to higher momentum states in quarkyonic matter and their chemical potentials rise accordingly. Furthermore, the quarkyonic mechanism can significantly stiffen the equation of state of hyperon star matter and thereby enhance the hyperon star maximum mass, thus helping to mitigate the hyperon puzzle.

nucl-th↗

arXiv2026

WEKP-PLP: A Wireless Environment Knowledge Pool-Enhanced Path Loss Prediction Framework for Shore-to-Ship Communication

Accurate shore-to-ship path loss prediction is essential for maritime mobile communication systems, but it remains challenging because dynamic sea surface conditions can alter reflection paths and multipath effects, introducing uncertainty into the observed path loss. This paper proposes a wireless environment knowledge pool-enhanced path loss prediction (WEKP-PLP) method for point prediction and interval characterization of shore-to-ship path loss. The proposed method first constructs a wireless environment knowledge pool (WEKP) from ray tracing (RT) simulations under different wind speeds, temperatures, salinities, frequencies, antenna heights, and propagation distances. The WEKP learns the mapping from environmental and system parameters to path loss residual quantiles and provides a prior that contains both the median prediction and the associated prediction interval. To adapt this prior to real scenarios, a small number of measurement samples are used to learn the residual between the WEKP median prediction and the measured path loss through Gaussian process regression (GPR). The experimental results show that the WEKP-PLP provides accurate point prediction and compact intervals with reliable coverage. The proposed method achieves an MAE of 1.03 dB and an RMSE of 1.54 dB. The ablation results further confirm that both the WEKP prior and the residual correction using a small number of measurement samples are necessary for accurate and reliable shore-to-ship path loss prediction.

eess.SP↗

arXiv2026

Support Thresholds, Not Algorithms, Limit Rare-Association Recovery in Co-Purchase Networks

The support threshold of the Apriori algorithm involves a trade-off in conducting market basket analysis: the associations that occur frequently are noted with high threshold; however, the low ones lead to generating the large amount of rules. The paper compares five methods for co-purchase edge filtration on two grocery datasets: i.e., Instacart (3.2 million baskets) and Dunnhumby (208 thousand baskets), including Apriori, Apriori + lift post-filtering, top-$K$ ranking based on lift, and two methods based on networks, noise-corrected (NC) and disparity filter (DF). The top-$K$ method ensures the maximum average lift, while the NC achieves similar lift level by means of a single value of the significance parameter ($α$). These two methods recover substantially more rare high-lift associations than Apriori (80-100% against 22-28%). NC and top-$K$ select meaningfully different edges (18-29% non-overlapping): NC retains statistically validated pairs, while top-$K$ retains rare pairs with high lift but low statistical significance. A rolling-origin holdout evaluation shows that top-$K$ edges recur at higher rates at every split, but NC edges are ~12 pp more likely to remain statistically significant in the held-out network.

cs.LG↗

arXiv2026

Boundary Embedding Shaping with Adaptive Contrastive Learning for Graph Structural Disentanglement

Graph neural networks (GNNs) excel at aggregating neighbor information for classification, yet their performance is hindered by graph structural entanglement, where spurious correlations from semantically irrelevant neighbors contaminate node embeddings. This challenge is most acute for nodes near class boundaries in the embedding space, where amplified structural noise blurs decision boundaries and destabilizes predictions. Existing robust GNN methods largely treat all nodes uniformly, ignoring boundary vulnerabilities. In this paper, to improve classification performance, we tackle graph structural disentanglement by identifying boundary-region entanglement as the primary bottleneck and propose Boundary Embedding Shaping (BES), an adaptive contrastive learning GNN plug-in module that selectively suppresses spurious structural noise at decision boundaries with minimal model parameter perturbation. Extensive experiments demonstrate that BES consistently improves boundary discrimination and outperforms existing leading methods. Notably, BES boosts GCN performance by an average of 3.3% in node classification (up to 5.0% on WikiCS) and achieves superior accuracy in link prediction.

cs.LG↗

arXiv2026

Length Value Model: Scalable Value Pretraining for Token-Level Length Modeling

Tokens are the fundamental units of computation in modern autoregressive models, and generation length directly influences both inference cost and reasoning performance. Despite its importance, existing approaches model length primarily at the coarse sequence level. We introduce the Length Value Model (LenVM), a token-level framework that estimates the remaining generation length at every decoding step. By formulating length modeling as a value estimation problem and assigning a constant negative reward to each generated token, LenVM predicts a bounded, discounted return that is a monotone proxy for the remaining generation horizon. This value formulation provides annotation-free, dense, unbiased, and scalable supervision. Experiments on LLMs and VLMs show that LenVM supports exact control, continuous performance--efficiency steering, length prediction, and interpretation. On LIFEBench-token, it raises the exact-length score of Qwen2.5-7B-Instruct from $30.9$ to $64.8$ in one pass and reaches $83.6$ when combined with LCG. Near $200$ tokens on GSM8K, LenVM retains about $63\%$ Pass@1 versus $6\%$ under a hard token budget. Across other settings, LenVM improves matched-length quality over prompting and EOS calibration, while its length-prediction accuracy improves with scale and exceeds specialized predictors. Its token-level values also identify markers of shifts between longer and shorter trajectories. These results establish generation length as an effective scalable token-level value signal for control, prediction, and future value-based training.

cs.CL↗