arXiv subjects

Lijuan Liu

Publications and source records attributed to Lijuan Liu.

At least 19 recordsLinked to original sources

arXiv2026

The coronal mass ejection productivity of solar active region 13664/8

In May 2024, NOAA active region (AR) complex 13664/8 was one of the most productive regions of the current solar cycle, producing 12 X-class flares and over 20 coronal mass ejections (CMEs) and triggering the strongest geomagnetic storm since 2003. We investigate why this AR complex was so CME-productive. Using primarily SDO/AIA and SDO/HMI observations, we analyzed its photospheric magnetic evolution, eruption sources, eruption waiting times, and magnetic parameters in comparison with five other ARs. The region initially contained only AR 13664 and exhibited limited flare activity until AR 13668 emerged on May 4, after which clustered major flares and CMEs occurred. Rapid, complex flux emergence substantially increased the region's area, magnetic flux, complexity, and nonpotentiality. The increased complexity was manifested by at least 12 emerging bipoles and six collisional polarity inversion lines (cPILs) formed between nonconjugated polarities, all showing sustained collision and shearing and serving as eruption sources. Decay index distributions show systematically lower critical heights (<45 Mm) above CME source regions. The CME waiting time distribution exhibits two peaks, suggesting that multiple cPILs enhanced CME productivity by increasing both recurrent CMEs from the same source and disturbance-triggered CMEs from nearby sources. These results indicate that, in addition to sufficient nonpotentiality and rapid background-field decay, high magnetic complexity accompanied by dynamical collisional shearing at multiple cPILs was crucial for the extreme CME productivity of AR 13664/8.

astro-ph.SR↗

arXiv2026

Learning Personalized Agents from Human Feedback

Modern AI agents are powerful but often fail to align with the idiosyncratic, evolving preferences of individual users. Prior approaches typically rely on static datasets, either training implicit preference models on interaction history or encoding user profiles in external memory. However, these approaches struggle with new users and with preferences that change over time. We introduce Personalized Agents from Human Feedback (PAHF), a framework for continual personalization in which agents learn online from live interaction using explicit per-user memory. PAHF operationalizes a three-step loop: (1) seeking pre-action clarification to resolve ambiguity, (2) grounding actions in preferences retrieved from memory, and (3) integrating post-action feedback to update memory when preferences drift. To evaluate this capability, we develop a four-phase protocol and two benchmarks in embodied manipulation and online shopping. These benchmarks quantify an agent's ability to learn initial preferences from scratch and subsequently adapt to persona shifts. Our theoretical analysis and empirical results show that integrating explicit memory with dual feedback channels is critical: PAHF learns substantially faster and consistently outperforms both no-memory and single-channel baselines, reducing initial personalization error and enabling rapid adaptation to preference shifts.

cs.AI↗

arXiv2025

Photospheric horizontal magnetic field decrease preceding a major solar eruption

Significant photospheric magnetic field changes during major solar eruptions -- interpreted as coronal feedback from eruptions to the photosphere -- are well-observed. However, analogous short-time scale field changes preceding eruptions are rarely reported. In this study, we present the first detailed analysis of a pre-flare decrease in the photospheric horizontal magnetic field ($B_h$) associated with an X1.8 class flare, using high-cadence vector magnetic field data from Helioseismic and Magnetic Imager onboard Solar Dynamics Observatory (SDO). We identify a region of gradual, spatially coherent $B_h$ decrease of about 100 G along the flaring polarity inversion line (PIL) during 30 minutes preceding the flare. This decrease is accompanied by a decrease in the force-free parameter $\alpha_w$, with no significant flux emergence or cancellation observed. After the flare onset, $B_h$ exhibited contrasting behaviors in different sub-regions: a step-like increase near the PIL and a continued decrease in surrounding regions, suggesting that the pre-flare $B_h$ decrease may also have a coronal origin, like its post-flare counterparts. Coronal imaging from Atmospheric Imaging Assembly onboard SDO reveals that the associated erupting filament underwent a slow-rise phase before the flare, whose timing and location closely matches the occurrence of the pre-flare $B_h$ decrease. We propose that the slow-rise of the pre-eruptive filament stretched overlying coronal loops, increasing their verticality and thereby reducing $B_h$ at their photospheric footpoints. The results present the first detailed analysis of a pre-flare $B_h$ decrease and suggest it as a precursor to solar eruptions, causally linked to early filament activation and its impact on the photosphere.

astro-ph.SR↗

arXiv2025

GeoWorld: Providing Full-frame Geometry Features to Facilitate 3D Scene Generation

Previous works that leverage video models for image-to-3D scene generation often suffer from geometric distortions and blurry content. Using video generation models to implicitly maintain geometric consistency according to a single-frame input is ineffective. In this paper, we present a two-stage method, named $\textbf{GeoWorld}$, that renovates the image-to-3D scene generation pipeline by providing full-frame geometry features. The first-stage video generation model, followed by a multi-view geometry model, produces $\textbf{full-frame}$ geometry features, which are then used as a mental draft of geometric conditions to aid the second-stage video-generation model. A geometric loss is proposed to impose real-world geometric constraints, and a geometry adaptation module is introduced to ensure the effective utilization of geometry features. Thanks to full-frame geometric modeling, the two smaller video models in our two-stage method can generate higher-fidelity 3D scenes than SOTA methods, while being even faster, e.g. 7.5$\times$ faster than Hunyuan-Voyager. Project page: https://peaes.github.io/GeoWorld.

cs.CV↗

arXiv2025

Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models

Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning and has become a standard post-training paradigm for contemporary language and vision-language models. However, the RLVR recipe introduces a significant risk of capability regression, in which models forget foundational skills after prolonged training without employing regularization strategies. We empirically confirm this concern, observing that open-source reasoning models suffer performance degradation on core capabilities such as perception and faithfulness. While imposing regularization terms like KL divergence can help prevent deviation from the base model, these terms are computed on the current task and therefore do not guarantee preservation of broader knowledge. Meanwhile, commonly used experience replay across heterogeneous domains makes it nontrivial to decide how much training emphasis each objective should receive. To address this, we propose RECAP-a replay strategy with dynamic objective reweighting for general knowledge preservation. Our reweighting mechanism adapts online using short-horizon signals of convergence and instability, shifting the post-training focus away from saturated objectives and toward underperforming or volatile ones. Our method is end-to-end and readily applicable to existing RLVR pipelines without training additional models or heavy tuning. Extensive experiments on benchmarks using Qwen2.5-VL-3B and Qwen2.5-VL-7B demonstrate the effectiveness of our method, which not only preserves general capabilities but also improves reasoning by enabling more flexible trade-offs among in-task rewards.

cs.LG↗

arXiv2025

Your thoughts tell who you are: Characterize the reasoning patterns of LRMs

Current comparisons of large reasoning models (LRMs) focus on macro-level statistics such as task accuracy or reasoning length. Whether different LRMs reason differently remains an open question. To address this gap, we introduce the LLM-proposed Open Taxonomy (LOT), a classification method that uses a generative language model to compare reasoning traces from two LRMs and articulate their distinctive features in words. LOT then models how these features predict the source LRM of a reasoning trace based on their empirical distributions across LRM outputs. Iterating this process over a dataset of reasoning traces yields a human-readable taxonomy that characterizes how models think. We apply LOT to compare the reasoning of 12 open-source LRMs on tasks in math, science, and coding. LOT identifies systematic differences in their thoughts, achieving 80-100% accuracy in distinguishing reasoning traces from LRMs that differ in scale, base model family, or objective domain. Beyond classification, LOT's natural-language taxonomy provides qualitative explanations of how LRMs think differently. Finally, in a case study, we link the reasoning differences to performance: aligning the reasoning style of smaller Qwen3 models with that of the largest Qwen3 during test time improves their accuracy on GPQA by 3.3-5.7%.

cs.CL↗

arXiv2025

IDCNet: Guided Video Diffusion for Metric-Consistent RGBD Scene Generation with Precise Camera Control

We present IDC-Net (Image-Depth Consistency Network), a novel framework designed to generate RGB-D video sequences under explicit camera trajectory control. Unlike approaches that treat RGB and depth generation separately, IDC-Net jointly synthesizes both RGB images and corresponding depth maps within a unified geometry-aware diffusion model. The joint learning framework strengthens spatial and geometric alignment across frames, enabling more precise camera control in the generated sequences. To support the training of this camera-conditioned model and ensure high geometric fidelity, we construct a camera-image-depth consistent dataset with metric-aligned RGB videos, depth maps, and accurate camera poses, which provides precise geometric supervision with notably improved inter-frame geometric consistency. Moreover, we introduce a geometry-aware transformer block that enables fine-grained camera control, enhancing control over the generated sequences. Extensive experiments show that IDC-Net achieves improvements over state-of-the-art approaches in both visual quality and geometric consistency of generated scene sequences. Notably, the generated RGB-D sequences can be directly feed for downstream 3D Scene reconstruction tasks without extra post-processing steps, showcasing the practical benefits of our joint learning framework. See more at https://idcnet-scene.github.io.

cs.CV↗

arXiv2025

Quality-aware Masked Diffusion Transformer for Enhanced Music Generation

Text-to-music (TTM) generation, which converts textual descriptions into audio, opens up innovative avenues for multimedia creation. Achieving high quality and diversity in this process demands extensive, high-quality data, which are often scarce in available datasets. Most open-source datasets frequently suffer from issues like low-quality waveforms and low text-audio consistency, hindering the advancement of music generation models. To address these challenges, we propose a novel quality-aware training paradigm for generating high-quality, high-musicality music from large-scale, quality-imbalanced datasets. Additionally, by leveraging unique properties in the latent space of musical signals, we adapt and implement a masked diffusion transformer (MDT) model for the TTM task, showcasing its capacity for quality control and enhanced musicality. Furthermore, we introduce a three-stage caption refinement approach to address low-quality captions' issue. Experiments show state-of-the-art (SOTA) performance on benchmark datasets including MusicCaps and the Song-Describer Dataset with both objective and subjective metrics. Demo audio samples are available at https://qa-mdt.github.io/, code and pretrained checkpoints are open-sourced at https://github.com/ivcylc/OpenMusic.

cs.SD↗

arXiv2025

High Accuracy, Less Talk (HALT): Reliable LLMs through Capability-Aligned Finetuning

Large Language Models (LLMs) currently respond to every prompt. However, they can produce incorrect answers when they lack knowledge or capability -- a problem known as hallucination. We instead propose post-training an LLM to generate content only when confident in its correctness and to otherwise (partially) abstain. Specifically, our method, HALT, produces capability-aligned post-training data that encodes what the model can and cannot reliably generate. We generate this data by splitting responses of the pretrained LLM into factual fragments (atomic statements or reasoning steps), and use ground truth information to identify incorrect fragments. We achieve capability-aligned finetuning responses by either removing incorrect fragments or replacing them with "Unsure from Here" -- according to a tunable threshold that allows practitioners to trade off response completeness and mean correctness of the response's fragments. We finetune four open-source models for biography writing, mathematics, coding, and medicine with HALT for three different trade-off thresholds. HALT effectively trades off response completeness for correctness, increasing the mean correctness of response fragments by 15% on average, while resulting in a 4% improvement in the F1 score (mean of completeness and correctness of the response) compared to the relevant baselines. By tuning HALT for highest correctness, we train a single reliable Llama3-70B model with correctness increased from 51% to 87% across all four domains while maintaining 53% of the response completeness achieved with standard finetuning.

cs.CL↗

arXiv2025

Diversity-driven Data Selection for Language Model Tuning through Sparse Autoencoder

Instruction tuning data are often quantity-saturated due to the large volume of data collection and fast model iteration, leaving data selection important but underexplored. Existing quality-driven data selection methods, such as LIMA (NeurIPS 2023 \citep{zhou2024lima}) and AlpaGasus (ICLR 2024 \citep{chenalpagasus}) generally ignore the equal importance of data diversity and complexity. In this work, we aim to design a diversity-aware data selection strategy and creatively propose using sparse autoencoders (SAEs) to tackle the challenge of data diversity measure. In addition, SAEs can also provide more interpretability of model behavior and explain, e.g., the surprising effectiveness of selecting the longest response (ICML 2024 \citep{zhaolong}). Using effective data selection, we experimentally prove that models trained on our selected data can outperform other methods in terms of model capabilities, reduce training cost, and potentially gain more control over model behaviors. We prove that SAEs can serve as a good alternative to diversity measure and design our method to be scalable for potential industrial large-scale pruning, and we will also release our trained SAEs for use by the broader community.

cs.CL↗

arXiv2024

Why could a new-born active region produce coronal mass ejections?

Solar active regions (ARs) are the main sources of flares and coronal mass ejections (CMEs). NOAA AR 12089, which emerged on 2014 June 10, produced two C-class flares accompanied by CMEs within five hours after its emergence. When producing the two eruptive flares, the total unsigned magnetic flux ($Φ_{\text{AR}}$) and magnetic free energy ($E_f$) of the AR are much smaller than the common CME-producing ARs. Why can this extremely small AR produce eruptive flares so early? We compare the AR magnetic environment for the eruptive flares to that for the largest confined flare from the AR. Besides the $Φ_{\text{AR}}$ and $E_f$, we calculate the ratio between the mean characteristic twist parameter ($α_{\text{FPIL}}$) within the flaring polarity inversion line (FPIL) region and $Φ_{\text{AR}}$, a parameter considering both background magnetic field constraint and non-potentiality of the core region, for the three flares. We find higher $α_{\text{FPIL}}/{Φ_{\text{AR}}}$ values during the eruptive flares than during the confined flare. Furthermore, we compute the decay index along the polarity inversion line, revealing values of 1.69, 3.45, and 0.98 before the two eruptive and the confined flares, respectively. Finally, nonlinear force-free field extrapolation indicates that a flux rope was repeatedly formed along the FPIL before eruptive flares, which ejected out and produced CMEs. No flux rope was found before the confined flare. Our research suggests that even a newly emerged, extremely small AR can produce eruptive flares if it has sufficiently weak background field constraint and strong non-potentiality in the core region.

astro-ph.SR↗

arXiv2024

Negligible Normal Fluid in Superconducting State of Heavily Overdoped Bi$_2$Sr$_2$CaCu$_2$O$_{8+δ}$ Detected by Ultra-Low Temperature Angle-Resolved Photoemission Spectroscopy

In high temperature cuprate superconductors, it was found that in the overdoped region the superfluid density decreases with the increase of hole doping. One natural question is whether there exists normal fluid in the superconducting state in the overdoped region. In this paper, we have carried out high-resolution ultra-low temperature laser-based angle-resolved photoemission measurements on a heavily overdoped Bi2212 sample with a $T_{\mathrm{c}}$ of 48 K. We find that this heavily overdoped Bi2212 remains in the strong coupling regime with $2 \mathitΔ_0 / k_{\mathrm{B}} T_{\mathrm{c}}=5.8$. The single-particle scattering rate is very small along the nodal direction ($\sim$5 meV) and increases as the momentum moves from the nodal to the antinodal regions. A hard superconducting gap opening is observed near the antinodal region with the spectral weight at the Fermi level fully suppressed to zero. The normal fluid is found to be negligibly small in the superconducting state of this heavily overdoped Bi2212. These results provide key information to understand the high $T_\mathrm{c}$ mechanism in the cuprate superconductors.

cond-mat.supr-con↗

arXiv2024

RedCoast: A Lightweight Tool to Automate Distributed Training of LLMs on Any GPU/TPUs

The recent progress of AI can be largely attributed to large language models (LLMs). However, their escalating memory requirements introduce challenges for machine learning (ML) researchers and engineers. Addressing this requires developers to partition a large model to distribute it across multiple GPUs or TPUs. This necessitates considerable coding and intricate configuration efforts with existing model parallel tools, such as Megatron-LM, DeepSpeed, and Alpa. These tools require users' expertise in machine learning systems (MLSys), creating a bottleneck in LLM development, particularly for developers without MLSys background. In this work, we present RedCoast (Redco), a lightweight and user-friendly tool crafted to automate distributed training and inference for LLMs, as well as to simplify ML pipeline development. The design of Redco emphasizes two key aspects. Firstly, to automate model parallelism, our study identifies two straightforward rules to generate tensor parallel strategies for any given LLM. Integrating these rules into Redco facilitates effortless distributed LLM training and inference, eliminating the need of additional coding or complex configurations. We demonstrate the effectiveness by applying Redco on a set of LLM architectures, such as GPT-J, LLaMA, T5, and OPT, up to the size of 66B. Secondly, we propose a mechanism that allows for the customization of diverse ML pipelines through the definition of merely three functions, avoiding redundant and formulaic code like multi-host related processing. This mechanism proves adaptable across a spectrum of ML algorithms, from foundational language modeling to complex algorithms like meta-learning and reinforcement learning. As a result, Redco implementations exhibit significantly fewer lines of code compared to their official counterparts.

cs.LG↗

arXiv2024

SMPLer: Taming Transformers for Monocular 3D Human Shape and Pose Estimation

Existing Transformers for monocular 3D human shape and pose estimation typically have a quadratic computation and memory complexity with respect to the feature length, which hinders the exploitation of fine-grained information in high-resolution features that is beneficial for accurate reconstruction. In this work, we propose an SMPL-based Transformer framework (SMPLer) to address this issue. SMPLer incorporates two key ingredients: a decoupled attention operation and an SMPL-based target representation, which allow effective utilization of high-resolution features in the Transformer. In addition, based on these two designs, we also introduce several novel modules including a multi-scale attention and a joint-aware attention to further boost the reconstruction performance. Extensive experiments demonstrate the effectiveness of SMPLer against existing 3D human shape and pose estimation methods both quantitatively and qualitatively. Notably, the proposed algorithm achieves an MPJPE of 45.2 mm on the Human3.6M dataset, improving upon Mesh Graphormer by more than 10% with fewer than one-third of the parameters. Code and pretrained models are available at https://github.com/xuxy09/SMPLer.

cs.CV↗

arXiv2023

Cappy: Outperforming and Boosting Large Multi-Task LMs with a Small Scorer

Large language models (LLMs) such as T0, FLAN, and OPT-IML, excel in multi-tasking under a unified instruction-following paradigm, where they also exhibit remarkable generalization abilities to unseen tasks. Despite their impressive performance, these LLMs, with sizes ranging from several billion to hundreds of billions of parameters, demand substantial computational resources, making their training and inference expensive and inefficient. Furthermore, adapting these models to downstream applications, particularly complex tasks, is often unfeasible due to the extensive hardware requirements for finetuning, even when utilizing parameter-efficient approaches such as prompt tuning. Additionally, the most powerful multi-task LLMs, such as OPT-IML-175B and FLAN-PaLM-540B, are not publicly accessible, severely limiting their customization potential. To address these challenges, we introduce a pretrained small scorer, Cappy, designed to enhance the performance and efficiency of multi-task LLMs. With merely 360 million parameters, Cappy functions either independently on classification tasks or serve as an auxiliary component for LLMs, boosting their performance. Moreover, Cappy enables efficiently integrating downstream supervision without requiring LLM finetuning nor the access to their parameters. Our experiments demonstrate that, when working independently on 11 language understanding tasks from PromptSource, Cappy outperforms LLMs that are several orders of magnitude larger. Besides, on 45 complex tasks from BIG-Bench, Cappy boosts the performance of the advanced multi-task LLM, FLAN-T5, by a large margin. Furthermore, Cappy is flexible to cooperate with other LLM adaptations, including finetuning and in-context learning, offering additional performance enhancement.

cs.LG↗

arXiv2023

Towards Garment Sewing Pattern Reconstruction from a Single Image

Garment sewing pattern represents the intrinsic rest shape of a garment, and is the core for many applications like fashion design, virtual try-on, and digital avatars. In this work, we explore the challenging problem of recovering garment sewing patterns from daily photos for augmenting these applications. To solve the problem, we first synthesize a versatile dataset, named SewFactory, which consists of around 1M images and ground-truth sewing patterns for model training and quantitative evaluation. SewFactory covers a wide range of human poses, body shapes, and sewing patterns, and possesses realistic appearances thanks to the proposed human texture synthesis network. Then, we propose a two-level Transformer network called Sewformer, which significantly improves the sewing pattern prediction performance. Extensive experiments demonstrate that the proposed framework is effective in recovering sewing patterns and well generalizes to casually-taken human photos. Code, dataset, and pre-trained models are available at: https://sewformer.github.io.

cs.CV↗

arXiv2023

The evolution of a spot-spot type solar active region which produced a major solar eruption

Solar active regions (ARs) are the main sources of large solar flares and coronal mass ejections. It is found that the ARs producing large eruptions usually show compact, highly-sheared polarity inversion lines (PILs). A scenario named as collisional-shearing is proposed to explain the formation of this type of PILs and the subsequent eruptions, which stresses the role of collision and shearing induced by relative motions of different bipoles in their emergence. However, in observations, if not considering the evolution stage of the ARs, about one third of the ARs that produce large solar eruptions govern a spot-spot type configuration. In this work, we studied the full evolution of an emerging AR, which owned a spot-spot type configuration when producing a major eruption, to explore the possible evolution gap between collisional shearing process in flux emergence and the formation of the spot-spot type, eruption-producing AR. It was found that the AR was formed through three bipoles emerged sequentially. The bipoles were arranged in parallel on the photosphere, so that the AR exhibited an overall large bipole configuration. In the fast emergence phase of the AR, the shearing gradually occurred due to the proper motions of the polarities, but no significant collision occurred due to the parallel arrangement of the bipoles. Nor did the large eruption occur. After the fast emergence, one large positive polarity started decay. Its dispersion led to the collision to a negative polarity which belonged to another bipole. A huge hot channel was formed through precursor flarings around the collision region. The hot channel erupted later, accompanied by an M7.3-class flare. The results suggest that in the spot-spot type AR, along with the shearing induced by the proper motions of the polarities, a decay process may lead to the collision of the polarities, driving the subsequent eruptions.

astro-ph.SR↗

arXiv2022

Nodal s$_\pm$ Pairing Symmetry in an Iron-Based Superconductor with only Hole Pockets

The origin of the high temperature superconductivity in the iron-based superconductors remains elusive after being extensively studied for more than a decade. Determination of the pairing symmetry is essential in understanding the superconductivity mechanism. In the iron-based superconductors that have hole pockets around the Brillouin zone center and electron pockets around the zone corners, the pairing symmetry is generally considered to be s$_\pm$, endowing a sign change in the superconducting gap between the hole and electron pockets. For the iron-based superconductors with only hole pockets, however, a couple of pairing scenarios have been proposed but the exact symmetry is still highly controversial. Here we report our determination of the pairing symmetry in KFe$_2$As$_2$ which is a prototypical iron-based superconductor with hole pockets both around the zone center and around the zone corners. By taking laser-based angle resolved photoemission measurements with super-high resolution and at ultra-low temperature, we have precisely determined the superconducting gap distribution and identified the locations of the gap nodes on all the Fermi surface around the zone center and the zone corners. The complete superconducting gap structure, in combination with the observation of the spin resonance in neutron scattering, provides strong evidence on the s$_\pm$ pairing symmetry in KFe$_2$As$_2$ with a gap sign reversal between the hole pockets around the zone center and the hole pockets around the zone corners. These results unify the pairing symmetry in the hole-doped iron-based superconductors and point to the spin fluctuation as the pairing glue in generating superconductivity.

cond-mat.supr-con↗