arXiv subjects

Yingjie Zhu

Publications and source records attributed to Yingjie Zhu.

At least 19 recordsLinked to original sources

arXiv2026

Revealing Magnon-Photon Coupling in Ultrathin Magnetic Films Using the Derivative-Divide Method

Magnon-photon coupling in planar cavities provides a promising platform for integrated cavity magnonics devices. However, its characterization becomes challenging in ultrathin magnetic films because their weak magnetic response can be obscured by the much stronger microwave background. Here, we employ the derivative-divide method to extract the field-dependent magnetic response from transmission measurements of split-ring resonators. The method reveals clear magn-photon anticrossings that are hardly discernible in conventional transmission spectra and enables quantitative extraction of the coupling strength. We resolve coupling in yttrium iron garnet films down to 60 nm and extend the approach to conductive CoFeB films down to 5 nm. Our results demonstrate a practical method for characterizing magno-photon coupling in ultrathin magnetic films, providing an experimental route toward cavity magnonics systems based on ultrathin magnetic films.

physics.app-ph↗

arXiv2026

Exploring if a coronal dimming event can produce coronal hole-like properties

Coronal dimmings, or transient coronal holes, are manifested as a sudden reduction in extreme ultraviolet (EUV) and X-ray emission, often following solar eruptions. We investigate whether a dimming event can produce coronal hole-like plasma characteristics by comparing imaging and spectroscopic observations from Solar Orbiter, Hinode, IRIS, and SDO prior and during the dimming. The SDO/AIA 193 Å emission intensity in the dimming region was reduced to the same level as the neighbouring coronal hole within 11 hours. The Doppler velocity measured with Fe XII (corona) decreased from $-0.34^{+0.37}_{-0.27}$ km/s towards a predominant upflow of $-3.2^{+0.4}_{-0.6}$ km/s. The first ionisation potential (FIP) bias was reduced towards photospheric values. We found an increase in the number of automatically detected EUV brightenings near the dimming boundary in SDO/AIA 193 Å which could be a sign of magnetic reconnection. In the cooler SDO/AIA 171 Å or Solar Orbiter HRI_EUV 174 Å channel, we did not observe such an increase. Coronal bright points (CBPs) appeared relatively unaffected by the formation of the dimming. The Mg II $k_3$ (chromosphere) Doppler velocities were unchanged, except for a small reduction in the dimming upflows in areas with weak magnetic fields ($<20$ G). We find that the dimming only shows partial coronal hole-like properties; specifically, in the coronal emission lines and at temperatures of $>1$ MK. We suggest that this is due to the dimming resulting from plasma depletion only at higher altitudes. Since the CBPs were not significantly impacted by the dimming, we used their magnetic loop heights ($\sim 10$ Mm) as the lower limit to the dimming height. Our findings provide new insights into the atmospheric structure of coronal dimmings.

astro-ph.SR↗

arXiv2026

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Existing methods typically evaluate unlearning effectiveness based on output deviations, while overlooking the generation quality after unlearning. This can easily lead to hallucinated or rigid responses, thereby affecting the usability and safety of the unlearned model. To address this issue, we propose ASRU, a controllable multimodal unlearning framework that incorporates generation quality as a core evaluation objective. ASRU first induces initial refusal behavior through activation redirection, and then optimizes fine-grained refusal boundaries using a customized reward function, thereby achieving a better trade-off between target knowledge unlearning and model utility. Experiments on Qwen3-VL show that ASRU significantly improves unlearning effectiveness (+24.6%) on average and generation quality (5.8X) on average while effectively preserving model utility, using only a small amount of retained supervision data.

cs.CL↗

arXiv2026

Decoupling Skeleton and Flesh: Efficient Multimodal Table Reasoning with Disentangled Alignment and Structure-aware Guidance

Reasoning over table images remains challenging for Large Vision-Language Models (LVLMs) due to complex layouts and tightly coupled structure-content information. Existing solutions often depend on expensive supervised training, reinforcement learning, or external tools, limiting efficiency and scalability. This work addresses a key question: how to adapt LVLMs to table reasoning with minimal annotation and no external tools? Specifically, we first introduce DiSCo, a Disentangled Structure-Content alignment framework that explicitly separates structural abstraction from semantic grounding during multimodal alignment, efficiently adapting LVLMs to tables structures. Building on DiSCo, we further present Table-GLS, a Global-to-Local Structure-guided reasoning framework that performs table reasoning via structured exploration and evidence-grounded inference. Extensive experiments across diverse benchmarks demonstrate that our framework efficiently enhances LVLM's table understanding and reasoning capabilities, particularly generalizing to unseen table structures. Our data and code are available at https://github.com/AAAndy-Zhu/TableVLM.

cs.CV↗

arXiv2026

Beyond the Vision Encoder: Identifying and Mitigating Spatial Bias in Large Vision-Language Models

Large Vision-Language Models (LVLMs) have achieved remarkable success across a wide range of multimodal tasks, yet their robustness to spatial variations remains insufficiently understood. In this work, we conduct a systematic study of the spatial bias of LVLMs, examining how models respond when identical key visual information is placed at different locations within an image. Through controlled probing experiments, we observe that current LVLMs often produce inconsistent outputs under such spatial shifts, revealing a clear spatial bias in their semantic understanding. Further analysis indicates that this bias does not stem from the vision encoder, but rather from a mismatch in attention mechanisms between the vision encoder and the large language model, which disrupts the global information flow. Motivated by this insight, we propose Adaptive Global Context Injection (AGCI), a lightweight mechanism that dynamically injects shared global visual context into each image token. AGCI works without architectural modifications, mitigating spatial bias by enhancing the semantic accessibility of image tokens while preserving the model's intrinsic capabilities. Extensive experiments demonstrate that AGCI not only enhances the spatial robustness of LVLMs, but also achieves strong performance on various downstream tasks and hallucination benchmarks.

cs.CV↗

arXiv2026

RelayGR: Scaling Long-Sequence Generative Recommendation via Cross-Stage Relay-Race Inference

Real-time recommender systems execute multi-stage cascades (retrieval, pre-processing, fine-grained ranking) under strict tail-latency SLOs, leaving only tens of milliseconds for ranking. Generative recommendation (GR) models can improve quality by consuming long user-behavior sequences, but in production their online sequence length is tightly capped by the ranking-stage P99 budget. We observe that the majority of GR tokens encode user behaviors that are independent of the item candidates, suggesting an opportunity to pre-infer a user-behavior prefix once and reuse it during ranking rather than recomputing it on the critical path. Realizing this idea at industrial scale is non-trivial: the prefix cache must survive across multiple pipeline stages before the final ranking instance is determined, the user population implies cache footprints far beyond a single device, and indiscriminate pre-inference would overload shared resources under high QPS. We present RelayGR, a production system that enables in-HBM relay-race inference for GR. RelayGR selectively pre-infers long-term user prefixes, keeps their KV caches resident in HBM over the request lifecycle, and ensures the subsequent ranking can consume them without remote fetches. RelayGR combines three techniques: 1) a sequence-aware trigger that admits only at-risk requests under a bounded cache footprint and pre-inference load, 2) an affinity-aware router that co-locates cache production and consumption by routing both the auxiliary pre-infer signal and the ranking request to the same instance, and 3) a memory-aware expander that uses server-local DRAM to capture short-term cross-request reuse while avoiding redundant reloads. We implement RelayGR on Huawei Ascend NPUs and evaluate it with real queries. Under a fixed P99 SLO, RelayGR supports up to 1.5$\times$ longer sequences and improves SLO-compliant throughput by up to 3.6$\times$.

cs.DC↗

arXiv2025

THCM-CAL: Temporal-Hierarchical Causal Modelling with Conformal Calibration for Clinical Risk Prediction

Automated clinical risk prediction from electronic health records (EHRs) demands modeling both structured diagnostic codes and unstructured narrative notes. However, most prior approaches either handle these modalities separately or rely on simplistic fusion strategies that ignore the directional, hierarchical causal interactions by which narrative observations precipitate diagnoses and propagate risk across admissions. In this paper, we propose THCM-CAL, a Temporal-Hierarchical Causal Model with Conformal Calibration. Our framework constructs a multimodal causal graph where nodes represent clinical entities from two modalities: Textual propositions extracted from notes and ICD codes mapped to textual descriptions. Through hierarchical causal discovery, THCM-CAL infers three clinically grounded interactions: intra-slice same-modality sequencing, intra-slice cross-modality triggers, and inter-slice risk propagation. To enhance prediction reliability, we extend conformal prediction to multi-label ICD coding, calibrating per-code confidence intervals under complex co-occurrences. Experimental results on MIMIC-III and MIMIC-IV demonstrate the superiority of THCM-CAL.

cs.CL↗

arXiv2025

Active region upflows in various coronal structures and their coupling to the lower atmosphere

Plasma upflows with a Doppler shift exceeding -10 km/s at active region (AR) boundaries are considered potential sources of the nascent slow solar wind. We investigate the driving mechanisms of a pair of coronal upflow regions on the western and eastern peripheries of an AR, which have different magnetic topologies and surroundings. It is aimed to explore how these upflows couple to the lower atmosphere. Using observations of the Fe XII 19.51 nm line from Hinode, we identified two upflow regions at the western and eastern boundaries of a decaying AR. Context images for the two regions were obtained by the High Resolution Imager (HRI) telescope of the Extreme Ultraviolet Imager (EUI) on board the Solar Orbiter mission. Other instruments on Solar Orbiter and other observatories provide diagnostics to the lower atmosphere. Potential Field Source Surface (PFSS) extrapolations were used to examine the magnetic field configuration associated with the AR upflows. The eastern upflow region, located over the AR moss, displays small-scale dynamic fibril structures, whereas the western region hosts fan-like loops. We found blueshifted Ne VIII emission at the eastern site, in contrast to redshifted Ne VIII profiles in the west. Magnetic field extrapolations reveal a pseudostreamer topology connecting both these regions. Moreover, low transition-region lines show systematically reduced redshift below the eastern footpoint. The observations support the scenario in which both upflows are driven by pressure imbalances created by coronal reconnection, leading to a continuous upflow above approximately 0.6 MK (i.e., Ne VIII line formation temperature). Meanwhile, mass flows in the lower transition region beneath the eastern upflow region appear to respond passively to the pressure-driven coronal upflows.

astro-ph.SR↗

arXiv2025

P/D-Device: Disaggregated Large Language Model between Cloud and Devices

Serving disaggregated large language models has been widely adopted in industrial practice for enhanced performance. However, too many tokens generated in decoding phase, i.e., occupying the resources for a long time, essentially hamper the cloud from achieving a higher throughput. Meanwhile, due to limited on-device resources, the time to first token (TTFT), i.e., the latency of prefill phase, increases dramatically with the growth on prompt length. In order to concur with such a bottleneck on resources, i.e., long occupation in cloud and limited on-device computing capacity, we propose to separate large language model between cloud and devices. That is, the cloud helps a portion of the content for each device, only in its prefill phase. Specifically, after receiving the first token from the cloud, decoupling with its own prefill, the device responds to the user immediately for a lower TTFT. Then, the following tokens from cloud are presented via a speed controller for smoothed TPOT (the time per output token), until the device catches up with the progress. On-device prefill is then amortized using received tokens while the resource usage in cloud is controlled. Moreover, during cloud prefill, the prompt can be refined, using those intermediate data already generated, to further speed up on-device inference. We implement such a scheme P/D-Device, and confirm its superiority over other alternatives. We further propose an algorithm to decide the best settings. Real-trace experiments show that TTFT decreases at least 60%, maximum TPOT is about tens of milliseconds, and cloud throughput increases by up to 15x.

cs.DC↗

arXiv2025

First coordinated observations between Solar Orbiter and the Daniel K. Inouye Solar Telescope

Solar Orbiter and the Daniel K. Inouye Solar Telescope (DKIST) are two of the newest facilities available to the solar physics community. The first coordinated observations of the Sun by these two facilities occurred over the course of one week in October 2022. The returned data are open-access and will provide a valuable resource to researchers in the field. We provide an overview of the datasets collected by Solar Orbiter and DKIST through this coordination and discuss their scientific potential. Our aim is to demonstrate how these unique high-resolution coordinated observations, as well as similar observations obtained through subsequent campaigns, can help tackle important science questions in the field. Between 18 and 24 October 2022, Solar Orbiter and DKIST observed a decayed active region simultaneously. During this period, Solar Orbiter's separation angle with Earth decreased from 77° to 51°, enabling stereoscopic observations with DKIST. From Solar Orbiter, observations are provided by the Extreme Ultraviolet Imager (EUI), Polarimetric and Helioseismic Imager (PHI), and the Spectral Imaging of the Coronal Environment (SPICE) instruments. Meanwhile, DKIST observed using the Cryogenic Near Infrared Spectropolarimeter (CryoNIRSP), the Visible Broadband Imager (VBI), and the Visible Spectropolarimeter (ViSP). The first coordinated Solar Orbiter and DKIST campaign was a success, collecting observations multiple times over the week. We focus on three specific topics as representative examples, namely, coronal loop physics, the formation and evolution of the small-scale active region brightenings, and coronal rain dynamics. These open-access observations, and others like them, should help the solar physics community tackle key questions in the field. Such stereoscopic coordinated observations open up a new era in the analysis of the solar atmosphere.

astro-ph.SR↗

arXiv2025

The Role of Excitatory Parvalbumin-positive Neurons in the Tectofugal Pathway of Pigeon (Columba livia) Hierarchical Visual Processing

The visual systems of birds and mammals exhibit remarkable organizational similarities: the dorsal ventricular ridge (DVR) demonstrates a columnar microcircuitry that parallels the cortical architecture observed in mammals. However, the specific neuronal subtypes involved and their functional roles in pigeon hierarchical visual processing remain unclear. This study investigates the role of excitatory parvalbumin (PV+) neurons within the Ento-MVL (entoallium-mesopallium venterolaterale) circuit of pigeons underlying hierarchical moving target recognition. Electrophysiological recordings and immunofluorescence staining reveal that excitatory PV+ neurons originating from the entopallial internal (Ei) predominantly modulate MVL responses to varying visual stimuli. Using a heterochronous-speed recurrent neural network (HS-RNN) model, we further validated these dynamics, replicating the rapid adaptation of the Ento-MVL circuit to moving visual targets. The findings suggest that the fast-spiking and excitatory properties of PV+ neurons enable rapid processing of motion-related information within the Ento-MVL circuit. Our results elucidate the functional role of excitatory PV+ neurons in hierarchical information processing under the columnar organization of the visual DVR and underscore the convergent neural processing strategies shared by avian and mammalian visual systems.

q-bio.NC↗

arXiv2025

SynGraph: A Dynamic Graph-LLM Synthesis Framework for Sparse Streaming User Sentiment Modeling

User reviews on e-commerce platforms exhibit dynamic sentiment patterns driven by temporal and contextual factors. Traditional sentiment analysis methods focus on static reviews, failing to capture the evolving temporal relationship between user sentiment rating and textual content. Sentiment analysis on streaming reviews addresses this limitation by modeling and predicting the temporal evolution of user sentiments. However, it suffers from data sparsity, manifesting in temporal, spatial, and combined forms. In this paper, we introduce SynGraph, a novel framework designed to address data sparsity in sentiment analysis on streaming reviews. SynGraph alleviates data sparsity by categorizing users into mid-tail, long-tail, and extreme scenarios and incorporating LLM-augmented enhancements within a dynamic graph-based structure. Experiments on real-world datasets demonstrate its effectiveness in addressing sparsity and improving sentiment modeling in streaming reviews.

cs.CL↗

arXiv2025

Benchmarking and Improving Large Vision-Language Models for Fundamental Visual Graph Understanding and Reasoning

Large Vision-Language Models (LVLMs) have demonstrated remarkable performance across diverse tasks. Despite great success, recent studies show that LVLMs encounter substantial limitations when engaging with visual graphs. To study the reason behind these limitations, we propose VGCure, a comprehensive benchmark covering 22 tasks for examining the fundamental graph understanding and reasoning capacities of LVLMs. Extensive evaluations conducted on 14 LVLMs reveal that LVLMs are weak in basic graph understanding and reasoning tasks, particularly those concerning relational or structurally complex information. Based on this observation, we propose a structure-aware fine-tuning framework to enhance LVLMs with structure learning abilities through three self-supervised learning tasks. Experiments validate the effectiveness of our method in improving LVLMs' performance on fundamental and downstream graph learning tasks, as well as enhancing their robustness against complex visual graphs.

cs.CL↗

arXiv2025

MiniMax-01: Scaling Foundation Models with Lightning Attention

We introduce MiniMax-01 series, including MiniMax-Text-01 and MiniMax-VL-01, which are comparable to top-tier models while offering superior capabilities in processing longer contexts. The core lies in lightning attention and its efficient scaling. To maximize computational capacity, we integrate it with Mixture of Experts (MoE), creating a model with 32 experts and 456 billion total parameters, of which 45.9 billion are activated for each token. We develop an optimized parallel strategy and highly efficient computation-communication overlap techniques for MoE and lightning attention. This approach enables us to conduct efficient training and inference on models with hundreds of billions of parameters across contexts spanning millions of tokens. The context window of MiniMax-Text-01 can reach up to 1 million tokens during training and extrapolate to 4 million tokens during inference at an affordable cost. Our vision-language model, MiniMax-VL-01 is built through continued training with 512 billion vision-language tokens. Experiments on both standard and in-house benchmarks show that our models match the performance of state-of-the-art models like GPT-4o and Claude-3.5-Sonnet while offering 20-32 times longer context window. We publicly release MiniMax-01 at https://github.com/MiniMax-AI.

cs.CL↗

arXiv2024

Dominance of 2-Minute Oscillations near the Alfvén Surface

Alfvén waves, considered one of the primary candidates for heating and accelerating the fast solar wind, are ubiquitous in spacecraft observations, yet their origin remains elusive. In this study, we analyze data from the first 19 encounters of the Parker Solar Probe (PSP) and report dominance of 2-minute oscillations near the Alfvén surface. The frequency-rectified trace magnetic power spectral density (PSD) of these oscillations indicates that the fluctuation energy is concentrated around 2 minutes for the ``youngest'' solar wind. Further analysis using wavelet spectrograms reveals that these oscillations primarily consist of outward-propagating, spherically polarized Alfvén wave bursts. Through Doppler analysis, we show that the wave frequency observed in the spacecraft frame can be mapped directly to the launch frequency at the base of the corona, where previous studies have identified a distinct peak around 2 minutes ($\sim 8$ mHz) in the spectrum of swaying motions of coronal structures observed by SDO AIA. These findings strongly suggest that the Alfvén waves originate from the solar atmosphere. Furthermore, statistical analysis of the PSD deformation beyond the Alfvén surface supports the idea of dynamic formation of the otherwise absent $1/f$ range in the solar wind turbulence spectrum.

astro-ph.SR↗

arXiv2024

CHECKWHY: Causal Fact Verification via Argument Structure

With the growing complexity of fact verification tasks, the concern with "thoughtful" reasoning capabilities is increasing. However, recent fact verification benchmarks mainly focus on checking a narrow scope of semantic factoids within claims and lack an explicit logical reasoning process. In this paper, we introduce CheckWhy, a challenging dataset tailored to a novel causal fact verification task: checking the truthfulness of the causal relation within claims through rigorous reasoning steps. CheckWhy consists of over 19K "why" claim-evidence-argument structure triplets with supports, refutes, and not enough info labels. Each argument structure is composed of connected evidence, representing the reasoning process that begins with foundational evidence and progresses toward claim establishment. Through extensive experiments on state-of-the-art models, we validate the importance of incorporating the argument structure for causal fact verification. Moreover, the automated and human evaluation of argument structure generation reveals the difficulty in producing satisfying argument structure by fine-tuned models or Chain-of-Thought prompted LLMs, leaving considerable room for future improvements.

cs.CL↗

arXiv2024

TSOM: Small Object Motion Detection Neural Network Inspired by Avian Visual Circuit

Detecting small moving objects in complex backgrounds from an overhead perspective is a highly challenging task for machine vision systems. As an inspiration from nature, the avian visual system is capable of processing motion information in various complex aerial scenes, and its Retina-OT-Rt visual circuit is highly sensitive to capturing the motion information of small objects from high altitudes. However, more needs to be done on small object motion detection algorithms based on the avian visual system. In this paper, we conducted mathematical modeling based on extensive studies of the biological mechanisms of the Retina-OT-Rt visual circuit. Based on this, we proposed a novel tectum small object motion detection neural network (TSOM). The neural network includes the retina, SGC dendritic, SGC Soma, and Rt layers, each layer corresponding to neurons in the visual pathway. The Retina layer is responsible for accurately projecting input content, the SGC dendritic layer perceives and encodes spatial-temporal information, the SGC Soma layer computes complex motion information and extracts small objects, and the Rt layer integrates and decodes motion information from multiple directions to determine the position of small objects. Extensive experiments on pigeon neurophysiological experiments and image sequence data showed that the TSOM is biologically interpretable and effective in extracting reliable small object motion features from complex high-altitude backgrounds.

cs.CV↗

arXiv2024

Spectroscopic Observations of the Solar Corona during the 2017 August 21 Total Solar Eclipse: Comparison of Spectral Line Widths and Doppler Shifts Between Open and Closed Magnetic Structures

The spectroscopic observations presented here were acquired during the 2017 August 21 total solar eclipse with a three-channel partially multiplexed imaging spectrometer (3PAMIS) operating at extremely high orders ($>$ 50). The 4 $R_\odot$ extent of the slit in the North-South direction scanned the corona starting from the central meridian out to approximately 1.0 $R_\odot$ off the east limb throughout totality. The line widths and Doppler shifts of the Fe X (637.4 nm) and Fe XIV (530.3 nm) emission lines, characteristic of $1.1 \times 10^6$ K and $1.8 \times 10^6$ K electron temperatures respectively, varied across the different coronal structures intercepted by the slit. Fe XIV was the dominant emission in the closed fields of an active region and the base of a streamer, with relatively constant 20 - 30 km s$^{-1}$ line widths independent of the height. In contrast, Fe X emission exhibited broader ($>40 $km s$^{-1}$) line widths in open fields which increased with height, in particular in the polar coronal hole. Inferences of line widths and Doppler shifts were consistent with extreme ultraviolet (EUV) observations from Hinode/EIS, as well as with the near-infrared Fe XIII 1074 nm line observed by CoMP. The differences in the spectral line widths between distinct coronal structures are interpreted as an indication of the predominance of wave heating in open structures versus localized heating in closed structures. This study underscores the unparalleled advantages and the enormous potential of TSE spectroscopy in measuring line widths simultaneously in open and closed fields at high altitudes, with minimal exposure times, stray light levels, and instrumental widths.

astro-ph.SR↗