Searcharxiv⌕ Search

arXiv subjects

Tingyu Qu

Publications and source records attributed to Tingyu Qu.

17 recordsLinked to original sources

Latent Bridge: Asking for Help Without Words

Text-based model collaboration requires a large model to generate intermediate responses before a smaller model can use its guidance, adding large-model decoding cost to inference. We introduce POND (Prefill-Only, No-Decode), a method that uses a frozen large sender for prefill and a small receiver for all autoregressive decoding. A trainable bridge maps the sender's hidden states to additional attention keys and values, allowing the receiver to use these representations throughout generation. POND also supports periodic review, in which the sender processes the receiver's partial response and updates its guidance through additional prefill passes while preserving the receiver's existing KV cache. Across reasoning and coding benchmarks, POND improves receiver performance with senders from different model families. In particular, GLM-5.3 provides gains comparable to those of Qwen3.8-Max when paired with a Qwen receiver, supporting the effectiveness of representation transfer between independently trained model families. POND further demonstrates its ability to generalize across domains. When trained on math and coding tasks, it transfers directly to general knowledge tasks without additional domain-specific training. These results demonstrate that large frontier model prefill representations can improve small-model generation without requiring large-model autoregressive decoding.

cs.CL↗

MobilePA-Bench: Benchmarking Mobile Planner Agents on Complex Real-World Tasks

Mobile planning requires a central agent to coordinate tools, persistent user context, reusable procedures, and specialized sub-agents while adapting to execution feedback. Evaluating these workflows requires distinguishing task completion from compliance with prescribed resource and procedural requirements. We introduce \textbf{MobilePA-Bench}, a benchmark comprising 1,705 tasks across eight query task domains and 212 mobile tools in a stateful simulated environment. It combines three features: (1) unified capability coverage, profiling Tool Use, Memory Usage, Skill Usage, and Sub-agent Collaboration under a common mobile protocol; (2) evidence-aligned verification, matching each task to a fixed tool-call, state-change, or interaction-behavior checker rather than imposing one completion criterion on all tasks; and (3) capability-gated evaluation, requiring task completion alongside prescribed memory checks and, in skill-only routing, complete route loading before business-tool use. With MobilePA-Bench, we systematically evaluate open- and closed-weight LLMs, revealing capability-specific weaknesses in complex mobile task execution. Fine-grained analyses distinguish failures in resource retrieval and procedural compliance from downstream execution errors, providing diagnostic insights beyond aggregate task-success rates. We further train a Qwen3.8-27B baseline with online reinforcement learning in the sandbox, whose improved performance provides initial evidence of the environment's utility for planner development.

cs.AI↗

Qwen-Planner-Agent: A Closed-Loop AI-for-AI Framework for Real-World Mobile Planner Agents

The rapid progression of large language models is extending AI from passive content generation into the active workflows of engineering and scientific discovery. This shift raises a compelling question: can AI be both the object of development and an active participant in building next-generation AI systems? We explore this question by building Qwen-Planner-Agent within a closed-loop AI-for-AI framework for scalable development and iterative improvement. Mobile planning offers a demanding test of this approach: complex, long-horizon tasks challenge agent reliability, while costly real-device interaction limits development scalability. The framework connects data production, model training, and deployment through a shared action-feedback-verification contract. (i) AI for Data builds a human-gated agentic data flywheel in which specialized agents construct tasks, collect interaction trajectories, curate and balance training data, and use training feedback to guide subsequent data generation. (ii) AI for Training combines a supervised planning cold start with hybrid-environment online agentic reinforcement learning, where we introduce Competence-Aware Reward-and-Advantage Engineering (CARE) to reduce reasoning and tool-use costs while preserving task performance. (iii) AI drives model--harness co-evolution through an execution-evidence-driven loop that orchestrates memory, skills, and tools at runtime and feeds structured action feedback and preserved failure traces back into coordinated model and harness adaptation. Qwen-Planner-Agent achieves the best overall performance among all evaluated models and systems on MobilePA-Bench, improving over its base model across tool use, memory, skills, and sub-agent coordination. Further evaluations of our model show improvements across non-mobile agentic benchmarks while largely preserving general capabilities.

cs.AI↗

Event Interaction in Low-Rank Bottlenecks for Temporal Relation Extraction

Temporal relation extraction determines whether an event occurs before, after, or simultaneously with another event, and therefore relies on accurately modeling how the two events interact. Mainstream systems achieve this by concatenating event spans or using shallow fusion, which works well when all model parameters are trainable. However, in parameter-efficient fine-tuning, low-rank bottlenecks restrict information flow and prevent these interaction signals from passing through, leading to clear performance drops. To address this limitation, we propose a theoretically grounded architecture, Convolutional Bottleneck Interaction (CBI), which first applies lightweight depthwise convolution to enhance event representations and then uses element-wise multiplication to capture effective event-event interactions inside the bottleneck. Across five datasets and seven backbone models in the Adapter and LoRA settings, CBI provides consistent and substantial gains, up to +31.7 micro F1, while adding minimal computational cost, showing that explicit interaction inside low-rank spaces is crucial for temporal relation extraction. The code is available at https://github.com/VRCMF/CIF.git.

cs.CL↗

OASIS: Online Sample Selection for Continual Visual Instruction Tuning

In continual instruction tuning (CIT) scenarios, where new instruction tuning data continuously arrive in an online streaming manner, training delays from large-scale data significantly hinder real-time adaptation. Data selection can mitigate this overhead, but existing strategies often rely on pretrained reference models, which are impractical in CIT setups since future data are unknown. Recent reference model-free online sample selection methods address this, but typically select a fixed number of samples per batch (e.g., top-k), making them vulnerable to distribution shifts where informativeness varies across batches. To address these limitations, we propose OASIS, an adaptive online sample selection approach for CIT that (1) selects informative samples by estimating each sample's informativeness relative to all previously seen data, beyond batch-level constraints, and (2) minimizes informative redundancy of selected samples through iterative selection score updates. Experiments on various large foundation models show that OASIS, using only 25 percent of the data, achieves comparable performance to full-data training and outperforms the state-of-the-art sampling methods.

cs.CV↗

Mitigating Negative Interference in Multilingual Sequential Knowledge Editing through Null-Space Constraints

Efficiently updating multilingual knowledge in large language models (LLMs), while preserving consistent factual representations across languages, remains a long-standing and unresolved challenge. While deploying separate editing systems for each language might seem viable, this approach incurs substantial costs due to the need to manage multiple models. A more efficient solution involves integrating knowledge updates across all languages into a unified model. However, performing sequential edits across languages often leads to destructive parameter interference, significantly degrading multilingual generalization and the accuracy of injected knowledge. To address this challenge, we propose LangEdit, a novel null-space constrained framework designed to precisely isolate language-specific knowledge updates. The core innovation of LangEdit lies in its ability to project parameter updates for each language onto the orthogonal complement of previous updated subspaces. This approach mathematically guarantees update independence while preserving multilingual generalization capabilities. We conduct a comprehensive evaluation across three model architectures, six languages, and four downstream tasks, demonstrating that LangEdit effectively mitigates parameter interference and outperforms existing state-of-the-art editing methods. Our results highlight its potential for enabling efficient and accurate multilingual knowledge updates in LLMs. The code is available at https://github.com/VRCMF/LangEdit.git.

cs.CL↗

Towards More Accurate Personalized Image Generation: Addressing Overfitting and Evaluation Bias

Personalized image generation via text prompts has great potential to improve daily life and professional work by facilitating the creation of customized visual content. The aim of image personalization is to create images based on a user-provided subject while maintaining both consistency of the subject and flexibility to accommodate various textual descriptions of that subject. However, current methods face challenges in ensuring fidelity to the text prompt while not overfitting to the training data. In this work, we introduce a novel training pipeline that incorporates an attractor to filter out distractions in training images, allowing the model to focus on learning an effective representation of the personalized subject. Moreover, current evaluation methods struggle due to the lack of a dedicated test set. The evaluation set-up typically relies on the training data of the personalization task to compute text-image and image-image similarity scores, which, while useful, tend to overestimate performance. Although human evaluations are commonly used as an alternative, they often suffer from bias and inconsistency. To address these issues, we curate a diverse and high-quality test set with well-designed prompts. With this new benchmark, automatic evaluation metrics can reliably assess model performance

cs.CV↗

Superlative spin transport of holes in ultra-thin black phosphorus

The development of energy-efficient spin-based hybrid devices that can perform functions such as logic, communication, and storage requires the ability to control and transport highly polarized spin currents over long distances in semiconductors. While traditional semiconductors such as silicon support spin transport, the effects of carrier type and concentration on important spin parameters are not well understood due to the need for extrinsic doping, which can cause additional momentum and hence spin scattering. Two-dimensional semiconductors, on the other hand, offer the ability to tune carrier type and concentration through field effect gating and inherently have long intrinsic spin lifetimes, making them a desirable platform for spin transport. Here, we study gate-tunable spin transport across narrow band-gap black phosphorus-based spin valves which enable us to systematically investigate spin transport with varying hole and electron concentrations under non-local geometry. Our findings demonstrate exceptional pure spin transport that approaches intrinsic limit, particularly in the low hole doping range. We achieved record non-local signals reaching 350 Ω and spin lifetimes exceeding 16 ns. Contrary to the behaviour seen in typical semiconductors, we find that the spin transport performance of holes in black phosphorus is significantly better than that of electrons, with the Elliott-Yafet process being the primary spin scattering mechanism. The observation of gate-tunable nanosecond spin lifetimes and colossal pure spin signals in both p- and n-type black phosphorus offers promising prospects for the development of novel semiconducting spintronics devices requiring sharp p-n interfaces.

cond-mat.mes-hall↗

TS-LLaVA: Constructing Visual Tokens through Thumbnail-and-Sampling for Training-Free Video Large Language Models

Recent advances in multimodal Large Language Models (LLMs) have shown great success in understanding multi-modal contents. For video understanding tasks, training-based video LLMs are difficult to build due to the scarcity of high-quality, curated video-text paired data. In contrast, paired image-text data are much easier to obtain, and there is substantial similarity between images and videos. Consequently, extending image LLMs for video understanding tasks presents an appealing alternative. Developing effective strategies for compressing visual tokens from multiple frames is a promising way to leverage the powerful pre-trained image LLM. In this work, we explore the limitations of the existing compression strategies for building a training-free video LLM. The findings lead to our method TS-LLaVA, which constructs visual tokens through a Thumbnail-and-Sampling strategy. Given a video, we select few equidistant frames from all input frames to construct a Thumbnail image as a detailed visual cue, complemented by Sampled visual tokens from all input frames. Our method establishes the new state-of-the-art performance among training-free video LLMs on various benchmarks. Notably, our 34B model outperforms GPT-4V on the MVBench benchmark, and achieves performance comparable to the 72B training-based video LLM, Video-LLaMA2, on the challenging MLVU benchmark. Code is available at https://github.com/tingyu215/TS-LLaVA.

cs.CV↗

A unified theoretical framework for Kondo superconductors: Periodic Anderson impurities with attractive pairing and Rashba spin-orbit coupling

Magnetic superconductors manifest a fascinating interplay between their magnetic and superconducting properties. This becomes evident, for example, in the significant enhancement of the upper critical field observed in uranium-based superconductors, or the destruction of superconductivity well below the superconducting transition temperature $T_c$ in cobalt-doped NbSe$_2$. In this work, we argue that the Kondo interaction plays a pivotal role in governing these behaviors. By employing a periodic Anderson model, we study the Kondo effect in superconductors with either singlet or triplet pairing. In the regime of small impurity energies and high doping concentrations, we find the emergence of a Kondo resistive region below $T_c$. While a magnetic field suppresses singlet superconductivity, it stabilizes triplet pairing through the screening of magnetic impurities, inducing reentrant superconductivity at high fields. Moreover, introducing an antisymmetric spin-orbital coupling suppresses triplet superconductivity. This framework provides a unified picture to understand the observation of Kondo effect in NbSe$_2$ as well as the phase diagrams in Kondo superconductors such as UTe$_2$, and URhGe.

cond-mat.supr-con↗

Introducing Routing Functions to Vision-Language Parameter-Efficient Fine-Tuning with Low-Rank Bottlenecks

Mainstream parameter-efficient fine-tuning (PEFT) methods, such as LoRA or Adapter, project a model's hidden states to a lower dimension, allowing pre-trained models to adapt to new data through this low-rank bottleneck. However, PEFT tasks involving multiple modalities, like vision-language (VL) tasks, require not only adaptation to new data but also learning the relationship between different modalities. Targeting at VL PEFT tasks, we propose a family of operations, called routing functions, to enhance VL alignment in the low-rank bottlenecks. These feature routing functions adopt linear operations and do not introduce new trainable parameters. In-depth analyses are conducted to study their behavior. In various VL PEFT settings, the routing functions significantly improve performance of the original PEFT methods, achieving over 20\% improvement on VQAv2 ($\text{RoBERTa}_{\text{large}}$+ViT-L/16) and 30\% on COCO Captioning (GPT2-medium+ViT-L/16). Also when fine-tuning a pre-trained multimodal model such as CLIP-BART, we observe smaller but consistent improvements across a range of VL PEFT tasks. Our code is available at https://github.com/tingyu215/Routing_VLPEFT.

cs.CV↗

Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps

Diffusion Probabilistic Models (DPM) have shown remarkable efficacy in the synthesis of high-quality images. However, their inference process characteristically requires numerous, potentially hundreds, of iterative steps, which could exaggerate the problem of exposure bias due to the training and inference discrepancy. Previous work has attempted to mitigate this issue by perturbing inputs during training, which consequently mandates the retraining of the DPM. In this work, we conduct a systematic study of exposure bias in DPM and, intriguingly, we find that the exposure bias could be alleviated with a novel sampling method that we propose, without retraining the model. We empirically and theoretically show that, during inference, for each backward time step $t$ and corresponding state $\hat{x}_t$, there might exist another time step $t_s$ which exhibits superior coupling with $\hat{x}_t$. Based on this finding, we introduce a sampling method named Time-Shift Sampler. Our framework can be seamlessly integrated to existing sampling algorithms, such as DDPM, DDIM and other high-order solvers, inducing merely minimal additional computations. Experimental results show our method brings significant and consistent improvements in FID scores on different datasets and sampling methods. For example, integrating Time-Shift Sampler to F-PNDM yields a FID=3.88, achieving 44.49\% improvements as compared to F-PNDM, on CIFAR-10 with 10 sampling steps, which is more performant than the vanilla DDIM with 100 sampling steps. Our code is available at https://github.com/Mingxiao-Li/TS-DPM.

cs.CV↗

Visually-Aware Context Modeling for News Image Captioning

News Image Captioning aims to create captions from news articles and images, emphasizing the connection between textual context and visual elements. Recognizing the significance of human faces in news images and the face-name co-occurrence pattern in existing datasets, we propose a face-naming module for learning better name embeddings. Apart from names, which can be directly linked to an image area (faces), news image captions mostly contain context information that can only be found in the article. We design a retrieval strategy using CLIP to retrieve sentences that are semantically close to the image, mimicking human thought process of linking articles to images. Furthermore, to tackle the problem of the imbalanced proportion of article context and image context in captions, we introduce a simple yet effective method Contrasting with Language Model backbone (CoLaM) to the training pipeline. We conduct extensive experiments to demonstrate the efficacy of our framework. We out-perform the previous state-of-the-art (without external data) by 7.97/5.80 CIDEr scores on GoodNews/NYTimes800k. Our code is available at https://github.com/tingyu215/VACNIC.

cs.CV↗

Evidence of the Coulomb gap in the density of states of MoS$_2$

$\mathrm{MoS_2}$ is an emergent van der Waals material that shows promising prospects in semiconductor industry and optoelectronic applications. However, its electronic properties are not yet fully understood. In particular, the nature of the insulating state at low carrier density deserves further investigation, as it is important for fundamental research and applications. In this study, we investigate the insulating state of a dual-gated exfoliated bilayer $\mathrm{MoS_2}$ field-effect transistor by performing magnetotransport experiments. We observe positive and non-saturating magnetoresistance, in a regime where only one band contributes to electron transport. At low electron density ($\sim 1.4\times 10^{12}~\mathrm{cm^{-2}}$) and a perpendicular magnetic field of 7 Tesla, the resistance exceeds by more than one order of magnitude the zero field resistance and exponentially drops with increasing temperature. We attribute this observation to strong electron localization. Both temperature and magnetic field dependence can, at least qualitatively, be described by the Efros-Shklovskii law, predicting the formation of a Coulomb gap in the density of states due to Coulomb interactions. However, the localization length obtained from fitting the temperature dependence exceeds by more than one order of magnitude the one obtained from the magnetic field dependence. We attribute this discrepancy to the presence of a nearby metallic gate, which provides electrostatic screening and thus reduces long-range Coulomb interactions. The result of our study suggests that the insulating state of $\mathrm{MoS_2}$ originates from a combination of disorder-driven electron localization and Coulomb interactions.

cond-mat.mes-hall↗

Ferromagnetic Superconductivity in Two-dimensional Niobium Diselenide

The co-existence of ferromagnetism and superconductivity becomes possible through unconventional pairing in the superconducting state. Such materials are exceedingly rare in solid-state systems but are promising platforms to explore topological phases, such as Majorana bound states. Theoretical investigations date back to the late 1950s, but only a few systems have so far been experimentally identified as potential hosts. Here, we show that atomically-thin niobium diselenide (NbSe$_2$) intercalated with dilute cobalt atoms spontaneously displays ferromagnetism below the superconducting transition temperature ($T_c$). We elucidate the origin of this phase by constructing a magnetic tunnel junction that consists of cobalt and cobalt-doped niobium diselenide (Co-NbSe$_2$) as the two ferromagnetic electrodes, with an ultra-thin boron nitride as the tunnelling barrier. At a temperature well below $T_c$, the tunnelling magnetoresistance shows a bistable state, suggesting a ferromagnetic order in Co-NbSe$_2$. We propose a RKKY exchange coupling mechanism based on the spin-triplet superconducting order parameter to mediate such ferromagnetism. We further perform non-local lateral spin valve measurements to confirm the origin of the ferromagnetism. The observation of Hanle precession signals show spin diffusion length up to micrometres below Tc, demonstrating an intrinsic spin-triplet nature in superconducting NbSe$_2$. Our discovery of superconductivity-mediated ferromagnetism opens the door to an alternative design of ferromagnetic superconductors

cond-mat.supr-con↗

Weakly Supervised Face Naming with Symmetry-Enhanced Contrastive Loss

We revisit the weakly supervised cross-modal face-name alignment task; that is, given an image and a caption, we label the faces in the image with the names occurring in the caption. Whereas past approaches have learned the latent alignment between names and faces by uncertainty reasoning over a set of images and their respective captions, in this paper, we rely on appropriate loss functions to learn the alignments in a neural network setting and propose SECLA and SECLA-B. SECLA is a Symmetry-Enhanced Contrastive Learning-based Alignment model that can effectively maximize the similarity scores between corresponding faces and names in a weakly supervised fashion. A variation of the model, SECLA-B, learns to align names and faces as humans do, that is, learning from easy to hard cases to further increase the performance of SECLA. More specifically, SECLA-B applies a two-stage learning framework: (1) Training the model on an easy subset with a few names and faces in each image-caption pair. (2) Leveraging the known pairs of names and faces from the easy cases using a bootstrapping strategy with additional loss to prevent forgetting and learning new alignments at the same time. We achieve state-of-the-art results for both the augmented Labeled Faces in the Wild dataset and the Celebrity Together dataset. In addition, we believe that our methods can be adapted to other multimodal news understanding tasks.

cs.CV↗

Heteromoiré Engineering on Magnetic Bloch Transport in Twisted Graphene Superlattices

Localized electrons subject to applied magnetic fields can restart to propagate freely through the lattice in delocalized magnetic Bloch states (MBSs) when the lattice periodicity is commensurate with the magnetic length. Twisted graphene superlattices with moiré wavelength tunability enable experimental access to the unique delocalization in a controllable fashion. Here we report the observation and characterization of high-temperature Brown-Zak (BZ) oscillations which come in two types, 1/B and B periodicity, originating from the generation of integer and fractional MBSs, in the twisted bilayer and trilayer graphene superlattices, respectively. Coexisting periodic-in-1/B oscillations assigned to different moiré wavelengths, are dramatically observed in small-angle twisted bilayer graphene, which may arise from angle-disorder-induced in-plane heteromoiré superlattices. Moreover, the vertical stacking of heteromoiré supercells in double-twisted trilayer graphene results in a mega-sized superlattice. The exotic superlattice contributes to the periodic-in-B oscillation and dominates the magnetic Bloch transport.

cond-mat.mes-hall↗