arXiv subjects

Ming Zhang

Publications and source records attributed to Ming Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

AI Can Learn Scientific Taste

Scientific discovery depends on expert judgement and foresight, which we call scientific taste: the ability to judge and propose research ideas with the potential for long-term scientific impact. Scientific taste is largely concentrated among highly experienced researchers, whose expertise is usually limited to a few specialised fields. If AI could learn scientific taste, it could reduce reliance on human experts and accelerate scientific discovery. Whether AI can learn this ability remains an open question. We introduce Reinforcement Learning from Community Feedback (RLCF) to learn judgement and ideation. Scientific Judge learns from community feedback, such as citations. Scientific Thinker learns to propose research ideas with high potential impact. Experiments show that Scientific Judge outperforms strong LLM baselines and that learned judgement generalises to future-year papers, other community metrics, and unseen fields. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than those proposed by baselines. These results suggest that AI can learn scientific taste, marking an important step towards AI systems that could help accelerate scientific discovery.

cs.CL↗

arXiv2026

Partial FC: Training 10 Million Identities on a Single Machine

Training face recognition models with millions of identities is challenging because classifier storage, logit memory, and computation grow linearly with the number of classes, eventually making full softmax impractical even when the backbone itself fits comfortably in memory. We present Partial FC (PFC), a scalable approximation to large-class softmax that preserves every positive class center while activating only a sampled subset of negative centers in each mini-batch. This asymmetric treatment retains every target term while avoiding exhaustive interaction with millions of mostly uninformative negatives. Our distributed implementation partitions the classifier across GPUs and samples within each owned shard, so sampling reduces local matrix multiplication and logit storage while sharding eliminates class-gradient synchronization across workers. Together, these properties reduce GPU-resident classifier memory, computation, and class-dependent communication without feature-based hard-negative retrieval. End-to-end system benchmarks demonstrate efficient scaling to massive class spaces, including 64 million classes on a single eight-GPU machine. Across large-scale face-recognition datasets, moderate sampling rates maintain competitive recognition accuracy while substantially improving training efficiency. Our best PFC configurations achieve 97.2\% TAR on IJB-C at FAR $=10^{-4}$ and 94.0\% TAR on ICCV21-MFR at FAR $=10^{-6}$. Beyond clean training data, PFC is robust to inter-class conflicts, label noise, and long-tailed identity distributions: under 40\% label noise, PFC with conflict filtering raises ICCV21-MFR TAR from 43.9\% to 80.2\%, while on long-tailed data PFC improves TAR from 87.4\% to 92.0\%. These results establish positive-preserving negative sampling as an effective foundation for scalable, accurate, and robust identity classification.

cs.CV↗

arXiv2026

Quasinormal frequencies and greybody factors for axial perturbations of dilaton-Euler-Heisenberg de Sitter black holes

We investigate the quasinormal modes (QNMs) and greybody factors of dilaton-Euler-Heisenberg (dEH) de Sitter (dS) black holes in string-inspired Euler-Heisenberg gravity. Since the axial gravitational and electromagnetic perturbations decouple, we treat them independently. By applying the asymptotic iteration method (AIM) alongside a sixth-order WKB approximation, we compute the quasinormal frequencies and find excellent agreement between the two approaches. We also find that the QNM spectra depend sensitively on the magnetic charge $Q_{\text{m}}$, cosmological constant $Λ$, and nonlinear coupling $ε$, with a notable topological anomaly appearing in the electromagnetic frequency trajectories. Additionally, larger values of $Q_{\text{m}}$ or the multipole number $l$ generally suppress wave transmission, while the electromagnetic sector with $ε=1$ exhibits an anomalous response.

hep-th↗

arXiv2026

Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning

We introduce Mobius-v0, an architecture that comprises a globally shared Memory (FFN) that stores knowledge vectors and multiple Reasoners (Self-Attn) that iteratively achieve compositional reasoning. Using hidden states as cache and carrier, reasoners repeatedly query memory for required knowledge-vectors, while the knowledge is transmitted back to reasoning operators. Through this knowledge-reasoning-separation architecture, Mobius achieves better knowledge compression and reasoning efficiency. Built upon Mobius-v0 architecture: 1) Our 7B model trained-from-scratch achieves similar downstream score as a 7B Transformer baseline with 62.6% of baseline's training data. 2) Our Intern-S2-Mobius, continually-pretrained from Qwen3.5-35B, achieves similar downstream score while delivering nearly 4x end-to-end inference speedup.

cs.AI↗

arXiv2026

MARCH: Scaling Recurrent Memory with Content-Routed State Anchors

Transformers owe much of their strong long-context retrieval capability to a token-level memory that grows with context length. This flexibility, however, incurs a quadratic computation complexity during training and a key--value cache that grows linearly during autoregressive inference. Recurrent alternatives offer efficient decoding by compressing the entire history into a fixed-size state, but often underperform on recall-intensive tasks since earlier associations usually get overwritten by subsequent updates, and only the most recent contextual information is retained. In this paper, we introduce Memory-Anchor Routing across Context History (MARCH), a network architecture that effectively scales state-space models beyond a fixed-size dimension, while maintaining computational efficiency over long-sequences. MARCH periodically caches cumulative recurrent-state checkpoints as state anchors and associates each anchor with a compact, content-conditioned anchor key. This lets MARCH maintain a memory bank, which can grow as context length increases, providing a controllable trade-off between historical resolution and memory cost. At each token, MARCH produces an anchor query to attend all causally available state anchors, and the output is calculated as an attention-style aggregation over all historical anchors along the current state. We show that after standard pretraining, MARCH consistently outperforms multiple linear attention variants across commonsense reasoning, LongBench, and in-context retrieval. These results demonstrate that content-routed state caching substantially strengthens recurrent long-range memory while preserving its native computation path.

cs.LG↗

arXiv2026

SPIEval: Evaluating Large Language Models as Mobile Assistants over Scattered Personal Information

Large language models (LLMs) are increasingly deployed as mobile assistants, where a key challenge is leveraging personal information scattered across multiple applications (apps) to complete user instructions. However, due to the lack of dedicated benchmarks, their capabilities remain poorly understood. To address this gap, we introduce SPIEval, a human-curated benchmark grounded in five cognitive capabilities (i.e., reasoning, disambiguation, integration, preference inference, and multi-intent decomposition). SPIEval comprises 250 tasks spanning 4,335 personal records distributed across 10 apps and supports multi-turn interaction through 21 tools. Analysis shows that the benchmark exhibits diverse scenarios, challenging tasks, scattered information, controllable environments, and verifiable outcomes. We evaluate nine representative LLMs and find substantial room for improvement. The best-performing model, GPT-5.5 (xhigh), achieves only 57.3% accuracy, while the weakest achieves just 16.4%. Further analysis reveals that 79% of failures stem from inaccurate information localization, as LLMs often commit to plausible but incorrect information instead of continuing retrieval for verification. We also find that fewer than 2% of retrieval actions employ advanced search methods and observe substantial variation in search efficiency across models. These findings expose fundamental limitations of current LLM-based mobile assistants and motivate future research in this direction. Data and code are available at https://huggingface.co/datasets/Junjie-Ye/SPIEval.

cs.CL↗

arXiv2026

Can Deep Research Agents Retrieve and Organize? Evaluating the Synthesis Gap with Expert Taxonomies

Deep Research Agents increasingly automate survey writing, yet existing benchmarks do not jointly test whether they retrieve the papers experts consider essential and organize those papers into paper-grounded taxonomies. We introduce TaxoBench, a benchmark built from 72 highly cited LLM surveys, 3,815 cited papers, and their expert-authored taxonomies. TaxoBench evaluates systems in two settings: Deep Research mode measures end-to-end retrieval and organization from a topic, while Bottom-Up mode provides the expert paper set and isolates organization. We evaluate leaf-level assignments with ARI and V-Measure and hierarchy-level structure with US-TED, US-NTED, and Sem-Path. Across 7 Deep Research Agents and 16 LLM configurations, the best agent retrieves only 20.92% of expert-cited papers, and none of 70 standard Bottom-Up runs reaches the experts' average depth of 4.86. A controlled probe shows that models which match this depth do so by fragmenting the taxonomy, reducing alignment with the expert reference. We further find that raw Sem-Path remains near a no-organization floor even when a newer model generation gains 3.68 pp ARI; after depth matching, humans lead on all 10 matched surveys by 13.27 pp. These results identify retrieval and hierarchical organization as separate bottlenecks and show why hierarchy metrics must be calibrated before they are used to compare models.

cs.CL↗

arXiv2026

Probing holographic conformal field theories

We embed relativistic quantum information protocols in AdS/CFT: an Unruh-DeWitt detector coupled to a local primary of a holographic CFT has a reduced state fixed by the universal boundary Wightman function. We find that the mana generated in a qutrit probe reads off the boundary condition of the dual bulk scalar, de-excitation spectroscopy tracks the double-trace flow through the lowest cylinder gap, and a local boundary detector is inequivalent to the HKLL representation of a bulk-local one.

hep-th↗

arXiv2026

PhySR: Physics-Informed Neural Network for Super-Resolution Reconstruction in Radio Synthesis Imaging

Radio telescope arrays are constrained by the number of antennas and baseline distribution, resulting in incomplete spatial-frequency sampling, limited image resolution, and blurring, distortion, and loss of small-scale structures caused by coupling between the primary and synthesized beams. Existing general-purpose model-driven methods remove observational effects sequentially and may accumulate errors, but cannot directly address limited imaging resolution, while data-driven methods lack explicit physical constraints. We propose PhySR, an end-to-end physics-informed neural network that combines a U-Net backbone, dynamic cascaded upsampling, a multiscale feature residual module, and a differentiable physical forward model incorporating the primary beam response, PSF convolution, and scale mapping. PhySR directly reconstructs high-resolution images from low-resolution dirty images without high-resolution labels while maintaining observation-domain consistency. Experiments on simulated SKA-Mid data show that, for 4x super-resolution, PhySR achieves a PSNR of 44.65 dB, an SSIM of 0.9940, and an RMSE of 0.0065. Compared with existing general-purpose methods, PSNR and SSIM improve by approximately 13.23 dB and 0.3760, respectively. Compared with mainstream deep learning models, PSNR and SSIM improve by 6.50 dB and 0.0682, while RMSE decreases by 0.0069. PhySR also remains stable for 2x and 8x super-resolution and achieves low observation-domain consistency errors, demonstrating advantages in coupling-effect removal, small-scale structure recovery, and physical consistency.

astro-ph.IM↗

arXiv2026

OxygenREC-v2: Internalizing Discrimination into Generative Recommendation

Generative recommendation unifies retrieval and ranking within a single model by autoregressively decoding semantic identifier (SID) sequences. Yet reliably incorporating behavior signals from clicks, cart additions, and orders remains challenging. Existing approaches either jointly optimize generative and discriminative objectives, requiring delicate trade-offs, or use a separate ranker as a post-hoc reinforcement-learning reward, risking out-of-distribution scoring and reward misalignment. We propose OxygenREC-v2, a generative recommender that Internalizes Discrimination into Generative Recommendation (IDGR). Rather than adding a separate discriminative objective, OxygenREC-v2 uses logged behavior to condition generation and supervise training. During pre-training, a behavior instruction conditions generation on the target behavior. During post-training, future interaction behaviors are exploited as privileged knowledge in our entropy-aware trajectory optimization self-distillation framework, enabling reward-model-free policy optimization. Throughout both training stages, OxygenREC-v2 maintains a single unified backbone. We implement OxygenREC-v2 as a 3B-parameter, 1B-activated MoE and deploy it on JD.com's large-scale e-commerce platform. Across multiple online A/B tests, OxygenREC-v2 improves user click-through conversion rate (UCTCVR) by 1.6--4.4% and GMV by 2.8--6.8% over OxygenREC-v1.

cs.IR↗

arXiv2026

Pairing mechanism and superconductivity in 1313 phase La$_3$Ni$_2$O$_7$

Recently, the observation of superconductivity (SC) with $T_c$ $\approx$ 3.6 K in the pressurized 1313 La$_3$Ni$_2$O$_7$ has attracted considerable interest. Here, we systematically investigate the electronic properties and superconducting mechanism of 1313 La$_3$Ni$_2$O$_7$ using density functional theory plus dynamical mean-field theory (DFT+DMFT) and random phase approximation (RPA). Our DFT+DMFT calculations reveal that the single-layer (SL) subsystem exhibits nearly insulating behavior, with the $d_{z^2}$ orbital showing Mott physics, while the trilayer (TL) subsystem remains metallic. This indicates that SC primarily resides in the TL subsystem, whose Ni-$e_g$ orbitals are found to be hole-doped relative to bulk La$_4$Ni$_3$O$_{10}$. Based on DFT+DMFT-derived low-energy Hamiltonian, RPA-based analysis yields an $s^{\pm}$-wave pairing symmetry within the TL subsystem. Importantly, we identify two key factors that contribute to the significant suppression of $T_c$ in 1313 La$_3$Ni$_2$O$_7$ compared to bulk La$_4$Ni$_3$O$_{10}$. First, the hole doping in the TL subsystem, as established by DMFT, leads to a decreased pairing strength, as confirmed by RPA calculations -- a trend resembling that in bulk La$_4$Ni$_3$O$_{10}$. Second, the SL subsystem acts as a bridge connecting adjacent superconducting TL subsystems, thereby forming an S-N-S Josephson junction. The resulting interlayer Josephson coupling governs the phase coherence between TL subsystems and further suppresses the global $T_c$. Combinedly, our findings suggest that the high-$T_c$ phase in the RP La$_3$Ni$_2$O$_7$ family should be attributed to the 2222 La$_3$Ni$_2$O$_7$ rather than the 1313 La$_3$Ni$_2$O$_7$.

cond-mat.supr-con↗

arXiv2026

Fully-connected three-mode squeezed vacuum: Gaussian entanglement, steering, and collective photon subtraction

We investigate a fully-connected three-mode squeezed vacuum (FC-C3MSV) state, where all three modes are pairwise coupled through nonlinear interactions in a triangle ($K_3$) topology. Using the integration-within-ordered-product technique, we derive the normal product form of the squeezing operator and obtain the covariance matrix directly from the Bogoliubov transformation. Under symmetric coupling, the physical state is genuinely tripartite entangled for any nonzero squeezing, while the three Armstrong-type witnesses provide a finite-window sufficient experimental test; in the chain-type C3MSV only one of these witnesses is violated. We find that, despite two-mode entanglement, the fully-connected topology admits \emph{no} two-mode Gaussian steering ($\mathcal{G}^{i\to j}=0$) between any pair of physical modes; the steering resource is instead collective one-mode-versus-two steering $\mathcal{G}^{i\to jk}$, which is $θ$-independent and grows with $r$. We analyze independent vacuum losses and obtain critical transmittances for steering survival: under full symmetric loss at $r=0.5$, one-to-two collective steering disappears at $η\approx0.58$, whereas reverse two-to-one collective steering survives down to $η\approx0.502$ and the underlying two-mode entanglement persists for all $η>0$. Finally, we revisit photon subtraction using a normalized phase-space derivation. A photon subtraction on a single physical mode does not generate Wigner negativity on another single mode, consistent with the absence of two-mode steering. Wigner negativity can instead be generated when Bob subtracts from the collective mode $(b+c)/\sqrt{2}$, with a loss threshold $η_c\approx0.667$ at $r=0.5$. These results distinguish pairwise and collective nonclassical resources in the FC-C3MSV and clarify the operational role of the complete-graph topology.

quant-ph↗

arXiv2026

Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget

We introduce Boogu-Image-0.1, an open-source unified multimodal understanding and generation model family, comprising Base, Turbo, Edit, and Edit-Turbo variants. It delivers competitive performance in high-quality text-to-image generation, fast inference, instruction-based editing, and bilingual (Chinese-English) text rendering. Closed-source multimodal systems like Nano-Banana-Pro and GPT-Image-2 achieve strong performance through system-level integration rather than a single model, yet their internal practices remain largely undisclosed. In this work, we demonstrate that strengthening the understanding capability of the system, through a stronger multimodal encoder, agentic prompt rewriting, and related techniques, together with improvements in data quality, training pipelines, and agentic inference-time scaling, can substantially enhance generation and editing performance even under highly constrained compute budgets. Comprehensive evaluations show that Boogu-Image-0.1 consistently matches or surpasses other open-source models across standard benchmarks, and achieves results approaching leading closed-source systems. Notably, this is accomplished with only 208.62 million unique images. The base model's theoretical training cost is only approximately \$400K. We share practical discussions that we believe are valuable to the broader research community, and release weights, code, and recipes under Apache 2.0 to advance the open ecosystem for unified multimodal understanding and generation. Our code is available here: https://github.com/Boogu-Project/Boogu-Image.

cs.CV↗

arXiv2026

Filling and Interlayer Superexchange Control Superconductivity in La$_3$Ni$_2$O$_7$

A central puzzle in bilayer nickelate superconductors is why pressure, epitaxial strain, oxygen stoichiometry, and chemical substitution produce systematic but apparently different changes in the superconducting T$_\text{c}$. Here we show that these trends can be organized by a two-parameter control principle based on the $d_{x^2-y^2}$-orbital filling $n_x$ and the effective interlayer antiferromagnetic superexchange $J_\perp$. Starting from a physical picture in which the nearly half-filled $d_{z^2}$ orbital supplies localized spin correlations while the nearly quarter-filled $d_{x^2-y^2}$ orbital carries superconductivity, we study an effective bilayer $t-J_\parallel-J_\perp$ model with parameters constrained by first-principles calculations. Combined slave-boson mean-field and density-matrix renormalization group calculations place realistic La$_3$Ni$_2$O$_7$ in an overdoped-cuprate-like regime where T$_\text{c}$ is governed mainly by the pairing scale. In this regime, hole doping reduces $n_x$ and suppresses T$_\text{c}$, whereas tuning routes that enhance $J_\perp$ raise T$_\text{c}$. This framework accounts for the suppression by over-oxidation and Ca/Sr substitution, the half-dome oxygen-stoichiometry response in the film, the enhancement by Nd/Sm substitution in the pressurized bulk, the right-triangle-like pressure dependence in the bulk, and the enhancement by compressive strain in the film. It also separates clean carrier doping from oxygen-vacancy tuning: clean electron doping mainly increases $n_x$, whereas oxygen vacancies weaken the apical-oxygen-mediated exchange path and introduce disorder. This leads to a falsifiable materials-design prediction: clean electron doping should enhance the pairing scale if introduced without oxygen vacancies or strong pair-breaking disorder.

cond-mat.supr-con↗

arXiv2026

Pitch-Angle Scattering of Cosmic Rays: Confronting Theory with Observations

Cosmic ray (CR) propagation is controlled by scattering in turbulent magnetic fields in space. In general, diffusive propagation is governed by pitch-angle diffusion in phase space. In this study, pitch-angle diffusion in the local interstellar medium (LISM) deduced from the analysis of the CR small-scale anisotropy data from the Tibet AS$γ$ experiment is compared with theoretical predictions. While it is difficult to reconcile the inferred LISM pitch angle diffusion coefficient with conventional theoretical results of particle scattering by Alfvénic turbulence, we find remarkable agreement with the predictions of particle scattering by quasi-slab fast modes whose properties are shaped by damping in the warm ionized medium. These findings offer direct evidence that CR scattering is predominantly governed by fast-mode turbulence. Furthermore, the comparison between experimental and theoretical results imposes strong constraints on plasma and magnetic field parameters within the local bubble, indicating that the LISM is in a low $β$ condition. The turbulence in the LISM should be compressible with an amplitude in the range of $0.3\lesssim δB/B_0 <1$.

astro-ph.HE↗

arXiv2026

Observational Study of Multi-wavelength Synergistic Effects in 3C 120

The energy dissipation and particle acceleration mechanisms within relativistic jets remain fundamental questions in active galactic nuclei (AGN) research. In this paper, we present a comprehensive 13-year (2012-2025) multi-wavelength study of the broad-line radio galaxy 3C 120, utilizing Fermi-LAT ($γ$-ray), ASAS-SN (optical), and high-resolution VLBA (15 GHz and 43 GHz) monitoring. Cross-correlation analyses reveal that $γ$-ray flares lead radio emission by $11.08_{-1.88}^{+4.03}$ months at 15 GHz and $8.27_{-5.55}^{+3.45}$ months at 43 GHz. This frequency-dependent temporal hierarchy positions the high-energy dissipation zone upstream of the radio core, corroborating the opacity-driven core-shift effect. By tracking the parsec-scale jet morphology during major $γ$-ray flaring epochs, we demonstrate that radio outbursts systematically coincide with compact core brightening, pronounced surges in polarized flux, abrupt electric vector position angle rotations, and the subsequent ejection of superluminal knots. Ultimately, our internal radio correlations suggest that jet dynamics are governed by a dual mechanism: long-term kinematic and flux baseline variations are geometrically modulated by a secular jet precession, while rapid, highly energetic polarimetric bursts are driven by short-lived internal shocks propagating down the jet channel.

astro-ph.GA↗

arXiv2026

VRPO: Rethinking Value Modeling for Robust RL under Noisy Supervision in LLM Post-Training

Reinforcement Learning (RL) in real-world environments often suffers from ambiguous or incomplete reward supervision, which undermines policy stability and generalization. Such noise may cause models to ignore key information or even collapse in advantage estimation. We find that a strong value model is essential for absorbing unstable signals and producing reliable advantages, offering denser and more robust supervision than the reward model. To better optimize noisy supervision, we propose VRPO, a framework that enhances value modeling for robust RL in LLM post-training. VRPO integrates (1) auxiliary losses guided by entropy and perplexity from a frozen language model, and (2) a variational information bottleneck, enabling the value model to filter noise and capture key words. This design allows the value model to correct noise rewards and generate more reliable advantage estimates, transforming it from a passive predictor into an active noise regulator. Experiments on multi-turn dialogue, math reasoning, and science QA with both rule-based and model-based rewards show that VRPO consistently outperforms baselines such as PPO and GRPO. Our work highlight the central role of the value model in Robust RL and provide a principled and practical approach to policy optimization under noisy supervision.

cs.LG↗

arXiv2026

Review of Machine Learning Models for Solar Energetic Particle Prediction

Solar energetic particle (SEP) events have attracted increasing attention due to their significant radiation hazards for aviation, spacecraft electronics, and human missions beyond Earth's magnetosphere. From a scientific perspective, SEP events are intriguing because they arise from a set of physical processes extending from the solar surface and corona through the heliosphere, offering insight into particle acceleration and transport mechanisms that are widely applicable across astrophysics. Therefore, advancing our ability to understand and predict SEP events is essential both for deepening our knowledge of such mechanisms and for safeguarding space technologies and exploration. Traditionally, researchers have modeled SEPs using physics-based simulations and empirical methods. More recently, machine learning (ML) has emerged as a new tool for understanding and predicting SEP events. The purpose of this manuscript is to review the currently available ML models for SEP prediction, identify the datasets used for training, compare their architectures, inputs, and outputs, and, based on these insights, outline good practices and recommendations for future research.

astro-ph.SR↗