Search arXiv⌕ Search

arXiv subjects

Zhen Zhang

Publications and source records attributed to Zhen Zhang.

At least 19 recordsLinked to original sources
arXiv2026

On the Value of Cross-Modal Misalignment in Multimodal Representation Learning

Multimodal representation learning, exemplified by multimodal contrastive learning (MMCL) using image-text pairs, aims to learn powerful representations by aligning cues across modalities. This approach relies on the core assumption that the exemplar image-text pairs constitute two representations of an identical concept. However, recent research has revealed that real-world datasets often exhibit \emph{cross-modal misalignment}. There are two distinct viewpoints on how to address this issue: one suggests mitigating the misalignment, and the other leveraging it. We seek here to reconcile these seemingly opposing perspectives and to provide a practical guide for practitioners. Using latent variable models, we thus formalize cross-modal misalignment by introducing two specific mechanisms: \emph{selection bias}, where some semantic variables are absent in the text, and \emph{perturbation bias}, where semantic variables are altered---both leading to misalignment in data pairs. Our theoretical analysis shows that, under suitable assumptions, population global minimizers of an alignment--entropy surrogate yield representations that depend only on the semantic variables unaffected by selection and perturbation biases and locally identify this subset almost surely. Extensive empirical studies on synthetic data and real image-text datasets further examine how cross-modal misalignment shapes semantic predictability and downstream performance, shedding light on its nuanced impact on multimodal representation learning.

cs.LG↗
arXiv2026

EP J062132.0+623837: A Multi-wavelength Study of a Changing-look Active Galactic Nucleus Discovered by Einstein Probe

EP J062132.0+623837 is a new X-ray source identified in 14-day binned survey data from the Wide-field X-ray Telescope aboard the Einstein Probe (EP). Subsequent X-ray observations with the Follow-up X-ray Telescope (FXT) on board EP localized the source to the central region of the galaxy 2MASX J06213207+6238390 at $z=0.084$. Here, we present a comprehensive analysis of the multi-wavelength data for this object. During the FXT and Swift/XRT follow-up, the source remained X-ray active, with observed luminosities of $(3-9)\times10^{43}$ erg s$^{-1}$, and its X-ray spectra were well described by power-law models with photon indices of $Γ\sim1.5$. The optical spectrum of the host galaxy clearly reveals a broad-line active galactic nucleus (AGN) exhibiting significantly enhanced broad Balmer emission lines compared to an archival spectrum obtained four years earlier, thereby marking a changing-look transition from Type 1.8 to Type 1.5. Furthermore, long-term optical and mid-infrared light curves indicate that the nucleus underwent dramatic brightening around 2010 and has since shown substantial variability. Given the spatial coincidence and multi-wavelength behavior, we identify EP J062132.0+623837 as a changing-look AGN, initially detected by EP and subsequently confirmed to exhibit large-amplitude variability across multiple bands. This case study demonstrates the great potential of wide-field time-domain X-ray surveys for discovering changing-look AGNs, paving a new avenue for their systematic identification as a powerful complement to ongoing optical surveys.

astro-ph.GA↗
arXiv2026

Right On Q: Recovering Reasoning under Quantization via Self-Generated Trajectories

Quantization-aware training (QAT) aims to recover reasoning capabilities in low-bit large language models (LLMs). Existing recipes typically supervise the quantized student with the original full-precision model on fixed response sequences. However, the resulting low-bit quantized models still struggle on long-horizon reasoning tasks and often fail to terminate within a token budget that suffices for the full-precision model. We identify two challenges in closing this gap. First, fixed-response supervision leaves a mismatch between training prefixes and those encountered during student generation, and this trajectory divergence grows as responses lengthen. Second, on its own prefixes, the student underpredicts some next-token candidates and overpredicts others, and this distributional misalignment becomes more frequent at high-entropy prefixes. We present OnQ, a low-bit recovery pipeline that combines on-policy distillation and candidate-wise Dynamic KL. Starting from block-wise calibration, the quantized student generates responses and receives full-precision teacher supervision on its own prefixes. Dynamic KL weights forward and reverse KL by each candidate's teacher-student probability ratio to balance correction of under- and overprediction. Across Qwen3-1.7B, 4B, and 8B at 2- and 3-bit precision, OnQ achieves the highest nine-benchmark average, with the largest gains on long-horizon reasoning benchmarks. At 2 bits, it outperforms every baseline on every benchmark at every model size, improving the Qwen3-8B average by 10.26 points over the strongest baseline. It also produces shorter responses with fewer truncations and repetitive errors.

cs.CL↗
arXiv2026

DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses

Scientific discovery requires developing worthwhile hypotheses into practical investigations, yet learning from this evolving research process remains challenging. We introduce an operational definition of valuable and testable research hypotheses: evidence-grounded claims about consequential uncertainties, coupled with feasible investigations, explicit decision criteria, and alternative explanations. Our DiscoveryRSI framework develops hypothesis trees through multi-agent evolutionary search, compiles selected investigations into executable Harbor tasks, and recursively learns through three mechanisms: hypothesis credit optimization attributes scientific assessments to documented research contributions; harness replay identifies productive strategies and recurring failures to refine research workflows; and rediscovery applies accepted improvements to new investigations, generating further experience for subsequent learning. We also introduce DFM-Bench, which provides comprehensive assessment of hypothesis quality and cumulative research progress at the levels of hypotheses, revisions, development paths and research runs. Across four life-science domains, automated evaluation shows that hypothesis quality rises from 4.86 for the base model to 5.39 after two rounds of self-improvement, with the large gains in falsifiability and data resolvability. We further validate hypothesis quality through expert review, comparing our DiscoveryRSI-27B with other research agents. Preliminary human assessments show that DiscoveryRSI-27B scores higher than GPT-6 Luna on hypothesis importance and whether claims are plausible and evidence-bounded. DiscoveryRSI-27B achieved a 6.3% rate of hypothesis correspondence with research published after the model cutoff, largely outperform GPT-6 with 2.5% and Claude Opus 5 with 2.3%, while delivering greater human-rated improvement along research paths.

cs.CL↗
arXiv2026

SPEM: Structure-Aware Procedural Memory for Efficient Skill Use in LLM Agents

LLM-based agents increasingly rely on reusable skills. However, scaling skill libraries makes efficient and accurate skill use difficult. Full skill loading incurs high token and runtime overhead. In contrast, flat semantic retrieval treats skills as independent text entries, ignoring dependencies, execution order, state transitions, and data flow constraints. We propose SPEM, a lightweight structure-aware procedural memory framework for efficient skill use in LLM-based agents. SPEM maintains Atomic Skill Memory for individual skills and Skill-DAG Memory for reusable compositions derived from successful executions. For a new task, SPEM reuses high-confidence Skill-DAG templates when available. Otherwise, it retrieves atomic skills and invokes Skill-GNN for state- and structure-aware skill selection. Successful executions are written back as new Skill-DAGs, allowing procedural memory to accumulate over time. Experiments on ALFWorld and SkillsBench show that SPEM achieves the best overall effectiveness--efficiency trade-off, reaching 87.1% ALFWorld success and 38.38% SkillsBench reward while reducing token and runtime costs against strong baselines.

cs.AI↗
arXiv2026

The Geometric Mechanics of Contrastive Representation Learning: Alignment Potentials, Entropic Dispersion, and Cross-modal Divergence

While InfoNCE underlies modern contrastive learning, its geometric mechanisms remain under-characterized beyond the canonical alignment--uniformity decomposition. We develop a measure-theoretic framework in which representation measures evolve on a fixed embedding manifold. In the large-batch limit, we prove value and gradient consistency, linking the stochastic objective to explicit deterministic energy landscapes and revealing a geometric bifurcation between unimodal and symmetric multimodal regimes. In the unimodal case, the intrinsic energy is strictly convex and admits a unique Gibbs equilibrium, showing that entropy acts as a tie-breaker within the aligned basin. In the multimodal case, the intrinsic geometry becomes cross-coupled and contains a persistent negative symmetric divergence term: each modality's marginal reshapes the effective landscape of the other, allowing strong pairwise alignment to coexist with a persistent modality gap. Controlled synthetic experiments and analyses of pretrained CLIP representations support these predictions. Overall, our results shift the analytical lens from pointwise discrimination to population geometry, showing that pairwise alignment alone is insufficient to control cross-modal marginal structure.

cs.LG↗
arXiv2026

Metric Poissonian Pair Correlation Property Part 2: The Case \(n\log n\)

We prove that $(αn\log n)_{n\ge1}$ has Poissonian pair correlation for almost every real $α$, solving a Problem posed by Aistleitner, El-Baz and Munsch [Open Problem 3, Geom. Funct. Anal. 2021]. Our main contribution is a new four-point counting estimate with a fixed power saving over $N^{7/3}$ at tolerance $N^{-1}$, valid on the ranges left after a first-moment and Borel--Cantelli argument. Its proof reduces the counting condition to the required proximity of an implicitly defined real root to an integer; a Fejér kernel majorizes this condition, and multidimensional van der Corput $A$- and $B$-processes supply the saving. A localized Mellin argument then yields the second-moment estimate needed for the metric conclusion.

math.NT↗
arXiv2026

Generalizable turbulence closures across bluff-body shapes by PINN-based solver-agnostic training

Data-driven turbulence closures are commonly calibrated by embedding a CFD solver within an inverse procedure that requires a converged flow solution at every optimization iteration, which is fragile and costly. We instead formulate closure identification as a PDE-constrained, data-driven inverse problem within a physics-informed neural network (PINN), and subsequently stabilize the inferred closure for deployment in a numerical solver. This separation eliminates the training cost of repeated CFD solves and associated adjoint or ensemble evaluations, providing substantial flexibility to investigate alternative closure forms. Using this flexibility, we develop four closures: three model the Reynolds stress on a realizable tensor basis---a local map, a non-local model transporting the turbulent kinetic energy and recovering the out-of-plane normal stress, and the same with a learned length scale $\ell$---and a fourth models the Reynolds force $\mathbf{F}=-\nabla\cdot\boldsymbolτ$ directly, free of the realizability constraint. All four are trained across six two-dimensional bluff-body wakes at $Re=10^4$ and deployed frozen in a standard finite-element solver, stabilized primarily by input-gradient smoothing. Under a strict leave-one-shape-out (LOSO) protocol, all four reduce the field errors of a steady SST $k$-$ω$ baseline, most strongly on the mean velocity. The learned-length-scale closure is most accurate on the stress fields, while the force model generalizes best on the mean velocity and drag (LOSO drag error $\sim 7\%$). The closures also train efficiently on Particle Image Velocimetry data, enabling geometries intractable for DNS.

physics.flu-dyn↗
arXiv2026

Ablation-Driven Selective Laser Direct Mass Transfer for Micro-LEDs with High Speed, Enhanced Gap Tolerance, and High Precision

Micro light-emitting diodes (micro-LEDs) offer superior performance for next-generation displays, yet their scalable manufacturing is severely bottlenecked by the mass transfer process from donor to receiver substrates. The existing laser-induced forward transfer (LIFT) methods are restricted by low throughput, limited selectivity, and long reaction times. In this study, we introduce a novel defocused direct laser ablation transfer (D-DLAT) method utilizing polyimide as a dynamic release layer. Whereas most indirect laser transfer methods trigger chip release followed by passive descent toward the receiver, D-DLAT directly ablates the release layer to actively propel the small-sized chips towards the receiver via ablation shock. By optimizing the ablation profiles and scanning paths, this method achieves high-speed transfer with significantly enhanced precision and robustness. We demonstrate the successful transfer of micro-LEDs ranging from 600 $\unicode{x03BC}$m down to 35 $\unicode{x03BC}$m in size, with a minimum reported placement error of less than 1 $\unicode{x03BC}$m, and a critical transfer gap of up to 50 $\unicode{x03BC}$m. Furthermore, the D-DLAT approach enables full selectivity, facilitating complex patterned transfers and array repitching operations that are essential for full-color display integration. Our findings indicate that D-DLAT effectively overcomes the fundamental limitations of current transfer technologies, demonstrating substantial potential for the scalable and high-yield manufacturing of advanced micro-LED displays. Ultimately, D-DLAT establishes a robust paradigm for the high-speed manipulation and heterogeneous integration of micro-optoelectronic devices, extending its profound implications beyond displays to the broader landscape of advanced micro-assembly and flexible electronics.

physics.optics↗
arXiv2026

Computed, not Memorized: Bit-Factorized Identifiers for Efficient Generative Retrieval

Generative retrieval is known for its efficient end-to-end indexing and retrieval. In particular, numeric Document Identifiers (DocIDs) are generally more efficient because they are shorter and generally fixed-length sequences over a small vocabulary, requiring fewer decoding steps during beam search. However, they tend to generalize poorly to unseen documents and must be learned during indexing-stage training. In this paper, we propose a numeric DocID design for generative retrieval that enables better adaptation to unseen documents without retraining. Our DocID design inherits the language model's knowledge: each DocID is a sequence of tokens factorized into bits, and each token is scored as the sum of its bits. Unlike existing numeric DocIDs, whose tokens carry meaning only through embeddings memorized during indexing training, each bit here derives its meaning from the model's representation. Therefore, the DocIDs of unseen documents remain scorable without retraining. Experiments on three benchmarks show that our DocID design substantially improves unseen-document retrieval while maintaining comparable performance on seen documents. It achieves over 20 percentage points higher Hit@10 on unseen documents and up to 7 times lower inference latency than existing numeric DocID designs on NQ320K and TriviaQA, while performing only 3 percentage points below weight-update methods on MS MARCO without additional training. These results show that an appropriate DocID design can heavily reduce retraining costs when adapting generative retrievers to new documents.

cs.IR↗
arXiv2026

Emergent medium-range order in amorphous silicon

Medium-range order (MRO) is an important feature of amorphous materials, but its nature in terms of atomic arrangement has remained elusive. Here, using large-scale machine-learning-driven atomistic simulations of amorphous silicon, we probe the MRO of this archetypal network glass. By characterizing the structure in three dimensions, we reveal that the symmetry of the atomic arrangement around a central atom shows a crossover as a function of distance. At short distances, the structure is determined by local bonding constraints, which evolve from predominantly tetrahedral environments in low-density amorphous silicon to highly coordinated and strongly disordered local environments in very-high-density amorphous silicon. At larger distances, however, distinct cubic- and octahedral-symmetry patterns emerge in the three-dimensional atomic-density distribution, despite the absence of these symmetries at shorter length scales. We find that this crossover is accompanied by a quasi-periodic spatial organization of the topological rings in the structure. The characteristic spacing between ring centers matches the length scale of the first sharp diffraction peak (FSDP) in the structure factor, establishing a direct connection between this reciprocal-space signature of MRO and its real-space organization. These results demonstrate that the FSDP, the spatial correlation of the rings, and emergent orientational symmetries are not independent manifestations of MRO, but reflect different aspects of a common structural organization. Our findings provide a unified real-space picture of MRO in amorphous silicon and suggest that intermediate-range structural order in network glasses emerges from the interplay between local bonding constraints and extended atomic arrangements driven by efficient packing.

cond-mat.soft↗
arXiv2026

Poissonian Pair Correlation for Bernoulli Orbits

Let \(α, β\in \mathbb{T} = \mathbb{R}/\mathbb{Z}\), and let \((ξ_n)_{n \ge 1}\) be independent random variables taking the values \(α\) and \(β\) with equal probability. We study the random walk \[X_n = ξ_1 + \cdots + ξ_n \pmod 1\] and present a deterministic Diophantine condition on \((α, β)\) that implies almost sure Poisson pair correlation. To the best of our knowledge, this is the first result providing explicit examples in this framework where the increment measure is purely discrete. The classical Fourier contraction method fails in this setting; a different approach is required. Our proof uses the parabolic geometry of returns to \(1\) and an exact covariance formula for overlapping blocks.

math.NT↗
arXiv2026

Memory-Augmented Phase-Field Modeling: Onsager Formulation, Internal-Variable Lifting, and Structure-Preserving Computation

In this work, we develop a general memory-augmented phase-field (MAPF) framework based on Onsager's variational principle, in which memory effects are incorporated through a general relaxation kernel while preserving the thermodynamic consistency of the underlying system. Rather than treating fractional models as fundamental formulations, we show that they can be interpreted as reduced descriptions associated with specific power-law memory kernels. The proposed MAPF equations provide a unified modeling framework for non-Markovian phase-field dynamics. To facilitate analysis and computation, we further introduce an equivalent internal-variable representation that transforms the original nonlocal-in-time dynamics into an extended Markovian gradient system. This formulation preserves key structural properties, including energy dissipation and mass conservation, and enables the development of efficient structure-preserving numerical schemes. In particular, we construct an energy-stable scalar auxiliary variable approach combined with a fast rank-one implementation that exploits the relaxation-spectrum-induced coupling structure, resulting in a computationally efficient algorithm with reduced storage requirements. The proposed framework establishes a systematic connection between memory-dependent phase-field modeling, internal-variable thermodynamics, and structure-preserving numerical computation. Numerical experiments demonstrate the accuracy, stability, and efficiency of the proposed methods for phase-field dynamics with long-time memory effects.

math.NA↗
arXiv2026

RLVR$^{2}$: Reinforcement Learning with Verifiable Rubric-based Ranking

Reinforcement Learning with Verifiable Rewards (RLVR) is expanding from tasks with well-defined correctness signals, such as mathematics and code, toward multifaceted quality requirements specified by multi-dimensional rubrics. Since policy optimization consumes one scalar per rollout, rubric-based pipelines must map multiple criterion scores into a scalar reward. This aggregation is often treated as score scaling, but it implicitly determines how quality dimensions trade off during training. The prevailing practice, normalizing each criterion and taking a linear combination, assumes that cardinal score differences are comparable across criteria and that gains on one criterion compensate for failures on another; both assumptions are unreliable when criteria are semantically heterogeneous. We propose Reinforcement Learning with Verifiable Rubric-based Ranking (RLVR$^2$), a verifiable ranking paradigm for rubric-based RLVR. For each criterion, RLVR$^2$ converts rubric scores into criterion-specific within-group ordinal outcomes, recovers a latent utility from the resulting comparison matrix, and merges these utilities into one training signal. By retaining only within-group ordering and discarding raw score magnitudes, RLVR$^2$ avoids calibrating heterogeneous rubric scales. It further supports objective-preserving attribute adjustment: auxiliary attributes that correlate with observed rankings but are not training objectives can enter the estimation without expanding the rubric or rewarding them directly. Across three model scales and 16 benchmarks, RLVR$^2$ consistently outperforms representative rubric-based baselines, achieving the best overall performance on most benchmarks at every scale. Analysis shows it controls systematic effects tied to reasoning efficiency and response formatting while preserving the quality objective.

cs.LG↗
arXiv2026

Understanding and Mitigating Training Instability in Large-Scale Recommender Systems: A Cascade Perspective

Training large-scale recommender systems on dynamic industrial platforms, such as Short-Form Video (SFV) services with billions of daily interactions, faces critical stability challenges. Rapid distribution shifts in user data and increasing model complexity frequently trigger a divergence cascade, typically initiating with gradient explosion, propagating to weight explosion, and culminating in loss explosion. This G-W-L cascade severely impedes model convergence and performance. This paper provides a systematic analysis of this G-W-L cascade within an industrial SFV recommender context. We then propose and evaluate a structured mitigation framework where interventions are explicitly mapped to counteract specific stages of the cascade: gradient (G) control, weight/activation (W) stabilization, and loss (L) shaping. Our contribution provides a structured framework for understanding and tackling training divergence, offering practical insights for deploying robust, continuously updated recommenders in large-scale, non-stationary settings. We empirically validate the effectiveness of these strategies on a major industrial SFV recommendation system.

cs.IR↗
arXiv2026

A nuclear clock synchronized to $^{229}$Th

Atomic clocks have made time and frequency the most precisely measured quantities in physics, progressing from microwave standards that realize the SI second to optical clocks with unprecedented precision. A nuclear clock transfers the frequency reference from an electronic to a nuclear transition, and the uniquely low-lying, laser-accessible isomeric transition in $^{229}$Th currently offers the most practical route to compact, robust timekeeping and sensitive tests of fundamental physics. Realizing such a clock requires turning spectroscopy of the $^{229}$Th nuclear resonance into a stable discriminator for steering a traceable oscillator. Here we demonstrate a $^{229}$Th nuclear clock by stabilizing a continuous-wave, narrow-linewidth 148.4-nm vacuum-ultraviolet (VUV) laser to a resolved, weakly temperature-sensitive nuclear transition in $^{229}$Th$:$CaF$_2$ crystals. A $10$-$μ\mathrm{W}$ VUV source generated by four-wave mixing in cadmium vapour and phototube-based frequency-modulation absorption readout provide a fast, high-signal-to-noise nuclear discriminator. The clock reaches a fractional frequency instability of $5\times10^{-13}/\sqrt{τ/\mathrm{s}}$ for averaging time $τ$. Clock-transition frequencies measured in two independently fabricated crystals agree at the $10^{-13}$ level and are consistent with previous VUV-comb measurements on other $^{229}$Th:CaF$_2$ crystals. These results establish laser-addressed nuclei as operational clock references and provide a reproducible solid-state platform for compact nuclear clocks, nuclear quantum sensors and precision tests of fundamental physics.

physics.atom-ph↗
arXiv2026

Understanding Affective Adaptation in Multimodal Foundation Models: Emergent Functional Specialization

Despite rapid progress in multimodal affective foundation models, how affective capabilities emerge within their internal architectures remains poorly understood. A critical open question is whether affective fine-tuning induces diffuse changes across the model or organizes computation into functionally specialized pathways. We systematically investigate this question through a broad module-level analysis of 13 affective model instances spanning nine model designs, multiple scales, tasks, and training paradigms, complemented by controlled functional analyses on representative models. We find that affective adaptation exhibits a consistent yet non-exclusive functional organization. Under matched trainable-parameter budgets, adapting only the feed-forward network (FFN) consistently outperforms adapting only the attention modules across all evaluated settings and, on average, nearly matches the performance obtained by tuning all major Transformer projections, identifying the FFN as a particularly efficient adaptation substrate. More strikingly, although the gate, up, and down projections exhibit comparable standalone adaptation capacity, their learned functional contributions become differentiated after joint optimization. Module recovery and targeted interventions identify \texttt{gate\_proj} as a particularly prominent pathway, while checkpoint analysis shows that this differentiation develops over the course of training. We characterize this phenomenon as emergent functional specialization: distinct pathway-level roles are not fully explained by standalone adaptation capacity, but arise through joint affective adaptation. Building on this finding, Gate-Focused Efficient Tuning (GET) retains 96.2-98.0\% of the performance obtained by tuning all major Transformer projections while using only 19.3-24.5\% as many trainable parameters.

cs.CV↗