arXiv subjects

Ting Cao

Publications and source records attributed to Ting Cao.

At least 55 records · Page 3Linked to original sources

arXiv2026

Learning to Commit: Generating Organic Pull Requests via Online Repository Memory

Large language model (LLM)-based coding agents achieve impressive results on controlled benchmarks yet routinely produce pull requests that real maintainers reject. The root cause is not functional incorrectness but a lack of organicity: generated code ignores project-specific conventions, duplicates functionality already provided by internal APIs, and violates implicit architectural constraints accumulated over years of development. Simply exposing an agent to the latest repository snapshot is not enough: the snapshot reveals the final state of the codebase, but not the repository-specific change patterns by which that state was reached. We introduce Learning to Commit, a framework that closes this gap through Online Repository Memory. Given a repository with a strict chronological split, the agent performs supervised contrastive reflection on earlier commits: it blindly attempts to resolve each historical issue, compares its prediction against the oracle diff, and distils the gap into a continuously growing set of skills-reusable patterns capturing coding style, internal API usage, and architectural invariants. When a new PR description arrives, the agent conditions its generation on these accumulated skills, producing changes grounded in the project's own evolution rather than generic pretraining priors. Evaluation is conducted on genuinely future, merged pull requests that could not have been seen during the skill-building phase, and spans multiple dimensions including functional correctness, code-style consistency, internal API reuse rate, and modified-region plausibility. Experiments on an expert-maintained repository with rich commit history show that Online Repository Memory effectively improves organicity scores on held-out future tasks.

cs.SE↗

arXiv2026

Robust valley-polarized excitonic Mott states and doublons enabled by stacking-controlled moiré geometry

Atomically-thin moiré superlattices offer an optically accessible platform for interacting bosons, where strong onsite repulsion $U_{xx}$ suppresses double occupancy and supports excitonic Mott states at unit filling. However, moiré confinement also enhances phonon- and disorder-assisted relaxation, challenging the robustness of these correlated states under dissipation. Here we show that strengthening the intersite exciton repulsion $V_{xx}$ between neighboring moiré cells offers a distinct route to stabilizing unit-filling excitonic Mott states. In H-stacked WSe2/WS2, moiré confinement endows interlayer excitons with an out-of-plane dipole and a pronounced in-plane quadrupolar charge distribution. Helicity-resolved transient photoluminescence, supported by first-principles-informed modelling, reveals that this quadrupolar geometry increases $V_{xx}$ at unit filling by at least a factor of two relative to the dipolar R-stacked excitons. Despite a slight reduction in $U_{xx}$, the enhanced $V_{xx}$ yields a long-lived, valley-polarized excitonic Mott state at unit filling that persists for ~12 ns - more than twice as long as in R-stacks - and remains robust up to ~50 K. Beyond unit filling, the same geometry supports valley-polarized doublons with fourfold longer lifetimes than in R-stacks. These results establish moiré-geometric control of intersite interactions as a route to stabilizing excitonic Mott states and doublons against dissipation in solids.

cond-mat.mes-hall↗

arXiv2026

Valleytronics in 2D Materials Roadmap

Valleytronics exploits non-equivalent energy extrema in the electronic band structure of crystalline solids -- the valley degree of freedom -- to encode, manipulate, and read out information. The advent of 2D materials, first graphene and then transition-metal dichalcogenides, made valley control practical through optical, electrical, and magnetic routes. This foundation has enabled remarkable progress in recent years spanning established frontiers, such as valley exciton physics and valley Hall effects, as well as emerging directions including lightwave valleytronics, nanophotonic integration, flat-band valleytronics, and spin-valley qubits. In parallel, there are sustained efforts to scale up valleytronic materials and to predict new valleytronic platforms. This Roadmap brings together perspectives from leading experts to chart the key opportunities and challenges at the forefront of 2D material valleytronics. Each section captures a snapshot of progress in a key research area, identifies critical open challenges, and outlines pathways toward future valleytronics breakthroughs.

cond-mat.mes-hall↗

arXiv2026

Advancing Mobile GUI Agents: A Verifier-Driven Approach to Practical Deployment

We propose V-Droid, a mobile GUI task automation agent. Unlike previous mobile agents that utilize Large Language Models (LLMs) as generators to directly generate actions at each step, V-Droid employs LLMs as verifiers to evaluate candidate actions before making final decisions. To realize this novel paradigm, we introduce a comprehensive framework for constructing verifier-driven mobile agents: the discretized action space construction coupled with the prefilling-only workflow to accelerate the verification process, the pair-wise progress preference training to significantly enhance the verifier's decision-making capabilities, and the scalable human-agent joint annotation scheme to efficiently collect the necessary data at scale. V-Droid obtains a substantial task success rate across several public mobile task automation benchmarks: 59.5% on AndroidWorld, 38.3% on AndroidLab, and 49% on MobileAgentBench, surpassing existing agents by 5.2%, 2.1%, and 9%, respectively. Furthermore, V-Droid achieves a remarkably low latency of 4.3s per step, which is 6.1x faster compared with existing mobile agents. The source code is available at https://github.com/V-Droid-Agent/V-Droid.

cs.AI↗

arXiv2026

Hummingbird: SLO-Oriented GPU Preemption at Microsecond-scale

Existing GPU-sharing techniques, including spatial and temporal sharing, aim to improve utilization but face challenges in simultaneously ensuring SLO adherence and maximizing efficiency due to the lack of fine-grained task scheduling on closed-source GPUs. This paper presents Hummingbird, an SLO-oriented GPU scheduling system that overcomes these challenges by enabling microsecond-scale preemption on closed-source GPUs while effectively harvesting idle GPU time slices. Comprehensive evaluations across diverse GPU architectures reveal that Hummingbird improves the SLO attainment of high-priority tasks by 9.7x and 3.5x compared to the state-of-the-art spatial and temporal-sharing approaches. When compared to executing exclusively, the SLO attainment of the high-priority task, collocating with low-priority tasks on Hummingbird, only drops by less than 1%. Meanwhile, the throughput of the low-priority task outperforms the state-of-the-art temporal-sharing approaches by 2.4x. Hummingbird demonstrates significant effectiveness in ensuring the SLO while enhancing GPU utilization.

cs.DC↗

arXiv2026

Atomically Resolved Acoustic Dynamics Coupled with Magnetic Order in a van der Waals Antiferromagnet

Magnetoelastic coupling in van der Waals (vdW) magnetic materials enables a unique interplay between the spin and lattice degrees of freedom. Characterizing the elastic responses with atomic and femtosecond resolution across the magnetic transition is essential for guiding the design of magnetically tunable actuators and strain-mediated spintronic devices. Here, ultrafast x-ray diffraction employed at a free-electron laser reveals that the atomic displacements, wave vectors, and dispersion relations of acoustic phonon modes in a vdW antiferromagnet FePS$_3$ are coupled with the magnetic order, by tracking both in-plane and out-of-plane Bragg peaks upon optical excitation across the Néel temperature (T$_N$). One transverse mode shows that a quasi-out-of-plane atomic displacement undergoes a significant directional change across T$_N$. Its quasi-in-plane wave vector is derived by the comparison between the measured sound velocity and the first-principles calculations. The other transverse mode is an interlayer shear acoustic mode whose amplitude is strongly enhanced in the antiferromagnetic phase, exhibiting eight times stronger amplitude than the longitudinal acoustic mode below T$_N$. The atomically resolved characterization of acoustic phonon dynamics that couple with magnetic ordering opens opportunities for harnessing unique magnetoelastic coupling in vdW magnets on ultrafast timescales.

cond-mat.mtrl-sci↗

arXiv2026

Strain-tunable magnetic correlations in spin liquid candidate Nb$_3$Cl$_8$

Recent research suggests the possibility of the two-dimensional breathing-Kagome magnet Nb$_3$Cl$_8$ hosting a quantum spin liquid state, warranting further study into its magnetic properties. Using ab initio calculations, we show that monolayer Nb$_3$Cl$_8$ has short-range antiferromagnetic correlations among Nb$_3$ trimers with S = 1/2, and becomes magnetically frustrated due to the underlying effective triangular lattice geometry, and is evidenced by a frustration index of f > 1. The high-temperature susceptibility shows a negative Weiss temperature from Monte Carlo calculations. Considering spin-orbit coupling, we investigate the magnetic anisotropy, including anisotropic exchange, single-ion anisotropy and the Dzyaloshinskii-Moriya interaction using the four-state energy mapping formalism. Although the elements have relatively small atomic numbers, the Dzyaloshinskii-Moriya interaction is comparable in magnitude to the anisotropic exchange. Additionally, we show that biaxial strain tunes the short-range correlations between antiferromagnetic, paramagnetic and ferromagnetic. These findings strengthen our understanding of Nb$_3$Cl$_8$ and advance its applications in current condensed matter physics and materials science research, including nanoscale mechanical and spintronics applications.

cond-mat.mes-hall↗

arXiv2026

BitDecoding: Unlocking Tensor Cores for Long-Context LLMs with Low-Bit KV Cache

The growth of long-context Large Language Models (LLMs) significantly increases memory and bandwidth pressure during autoregressive decoding due to the expanding Key-Value (KV) cache. While accuracy-preserving KV-cache quantization (e.g., 4-bit or 2-bit) reduces memory footprint, existing systems decode inefficiently by relying solely on CUDA cores, underutilizing Tensor Cores-the dominant compute resource on GPUs. We present BitDecoding, the first inference system to efficiently decode low-bit KV caches by cooperatively leveraging CUDA cores and Tensor Cores. BitDecoding smartly induces Tensor-Core-friendly layouts, introduces warp-level dequantization parallelism, and provides unified system support through query transformation, high-performance tensor- and channel-wise quantization, and a software-pipelined dequantization kernel enabling mixed-precision execution. Architecture-aware optimizations further leverage Hopper's warpgroup tensor instructions and Blackwell's NVFP4 (MXFP4) tensor formats. Evaluated on Blackwell, Hopper, and Ampere GPUs, BitDecoding achieves an average 7.5x decoding speedup over FP16 FlashDecoding-v2, up to 8.6x on Blackwell with NVFP4, and up to 4.3x over state-of-the-art approaches. On LLaMA-3.1-8B with a 128K context, BitDecoding reduces single-batch decoding latency by 3x. BitDecoding is open-sourced at https://github.com/OpenBitSys/BitDecoding.

cs.AR↗

arXiv2025

Zoomer: Adaptive Image Focus Optimization for Black-box MLLM

Multimodal large language models (MLLMs) such as GPT-4o, Gemini Pro, and Claude 3.5 have enabled unified reasoning over text and visual inputs, yet they often hallucinate in real world scenarios especially when small objects or fine spatial context are involved. We pinpoint two core causes of this failure: the absence of region-adaptive attention and inflexible token budgets that force uniform downsampling, leading to critical information loss. To overcome these limitations, we introduce Zoomer, a visual prompting framework that delivers token-efficient, detail-preserving image representations for black-box MLLMs. Zoomer integrates (1) a prompt-aware emphasis module to highlight semantically relevant regions, (2) a spatial-preserving orchestration schema to maintain object relationships, and (3) a budget-aware strategy to adaptively allocate tokens between global context and local details. Extensive experiments on nine benchmarks and three commercial MLLMs demonstrate that Zoomer boosts accuracy by up to 27% while cutting image token usage by up to 67%. Our approach establishes a principled methodology for robust, resource-aware multimodal understanding in settings where model internals are inaccessible.

cs.CV↗

arXiv2025

Layerwise Stratification and Band Reordering in Twisted Multilayer MoTe$_2$

We introduce a generalizable, physics informed strategy for generating training data that enables a machine learning force field accurate over a broad range of twist angles and stacking layer numbers in moire systems. Applying this to multilayer twisted MoTe2 (tMoTe2), we identify a structural and electronic stratification: the two moire interface (MI) layers retain substantial lattice reconstruction even in thick multilayers, while outer bulk like layers show rapidly attenuated distortions.Surprisingly, this stratification becomes strongest not in the ultra-small twist angle regime (<~1°), where in plane domain formation is well known, but rather at intermediate angles (2-5°). Simultaneously, interlayer hybridization across the MI-bulk boundary is strongly suppressed, leading to electronic isolation. In twisted double bilayer MoTe2, this stratification gives rise to coexisting honeycomb and triangular lattice motifs in the frontier valence bands. We further demonstrate that twist angle and weak gating can create energy shift of bands belonging to the two motifs, producing Chern band reordering and nonlinear electric polarization with modest hole doping. Our approach allows efficient simulation of multilayer moire systems and reveals structural-electronic separation phenomena absent in bilayer systems.

cond-mat.mtrl-sci↗

arXiv2025

T-MAN: Enabling End-to-End Low-Bit LLM Inference on NPUs via Unified Table Lookup

Large language models (LLMs) are increasingly deployed on customer devices. To support them, current devices are adopting SoCs (System on Chip) with NPUs (Neural Processing Unit) installed. Although high performance is expected, LLM inference on NPUs is slower than its CPU counterpart. The reason is that NPUs have poor performance on computations other than GEMM, like dequantization. Current works either disaggregate prefill on the NPUs and decoding on the CPUs, or put both on the NPUs but with an accuracy loss. To solve this issue, based on the insight that low-bit can enable target computation encoded within an acceptably sized table, we propose table lookup to subsume hardware operations otherwise unsupported. To realize this, we overcome the conflicting hardware behavior of prefill and decoding to design a unified table layout and tiling through (1) fused two-level table-based dequantization and (2) concurrency-hierarchy-guided tiling. Based on that, we implement the prefill phase by three-stage pipeline and map the table-lookup-based decoding to NPU's vector units. Results show 1.4x and 3.1x speedup for prefill and decoding respectively, and 84% energy savings compared to the baseline NPU methods. The code is available at https://github.com/microsoft/T-MAC/tree/main/t-man.

cs.AR↗

arXiv2025

AVA: Towards Agentic Video Analytics with Vision Language Models

AI-driven video analytics has become increasingly important across diverse domains. However, existing systems are often constrained to specific, predefined tasks, limiting their adaptability in open-ended analytical scenarios. The recent emergence of Vision Language Models (VLMs) as transformative technologies offers significant potential for enabling open-ended video understanding, reasoning, and analytics. Nevertheless, their limited context windows present challenges when processing ultra-long video content, which is prevalent in real-world applications. To address this, we introduce AVA, a VLM-powered system designed for open-ended, advanced video analytics. AVA incorporates two key innovations: (1) the near real-time construction of Event Knowledge Graphs (EKGs) for efficient indexing of long or continuous video streams, and (2) an agentic retrieval-generation mechanism that leverages EKGs to handle complex and diverse queries. Comprehensive evaluations on public benchmarks, LVBench and VideoMME-Long, demonstrate that AVA achieves state-of-the-art performance, attaining 62.3% and 64.1% accuracy, respectively-significantly surpassing existing VLM and video Retrieval-Augmented Generation (RAG) systems. Furthermore, to evaluate video analytics in ultra-long and open-world video scenarios, we introduce a new benchmark, AVA-100. This benchmark comprises 8 videos, each exceeding 10 hours in duration, along with 120 manually annotated, diverse, and complex question-answer pairs. On AVA-100, AVA achieves top-tier performance with an accuracy of 75.8%. The source code of AVA is available at https://github.com/I-ESC/Project-Ava. The AVA-100 benchmark can be accessed at https://huggingface.co/datasets/iesc/Ava-100.

cs.CV↗

arXiv2025

Long Exposure: Accelerating Parameter-Efficient Fine-Tuning for LLMs under Shadowy Sparsity

The adaptation of pre-trained large language models (LLMs) to diverse downstream tasks via fine-tuning is critical for numerous applications. However, the inefficiency of parameter-efficient fine-tuning (PEFT) techniques presents significant challenges in terms of time investments and operational costs. In this paper, we first introduce a nuanced form of sparsity, termed Shadowy Sparsity, which is distinctive in fine-tuning and has not been adequately addressed for acceleration. Under Shadowy Sparsity, we propose Long Exposure, an efficient system to accelerate PEFT for LLMs. Long Exposure comprises three key components: Shadowy-sparsity Exposer employs a prolonged sensing range to capture more sparsity details under shadowy sparsity; Sequence-oriented Predictor provides efficient yet accurate predictions to handle large sequence inputs and constantly-evolving parameters; and Dynamic-aware Operator facilitates more structured computational patterns and coalesced memory accesses, addressing dynamic sparse operations. Extensive evaluations show that Long Exposure outperforms state-of-the-arts with up to a $2.49\times$ speedup in end-to-end fine-tuning, offering promising advancements in accelerating PEFT for LLMs.

cs.LG↗

arXiv2025

Neuralink: Fast LLM Inference on Smartphones with Neuron Co-Activation Linking

Large Language Models (LLMs) have achieved remarkable success across various domains, yet deploying them on mobile devices remains an arduous challenge due to their extensive computational and memory demands. While lightweight LLMs have been developed to fit mobile environments, they suffer from degraded model accuracy. In contrast, sparsity-based techniques minimize DRAM usage by selectively transferring only relevant neurons to DRAM while retaining the full model in external storage, such as flash. However, such approaches are critically limited by numerous I/O operations, particularly on smartphones with severe IOPS constraints. In this paper, we propose Neuralink, a novel approach that accelerates LLM inference on smartphones by optimizing neuron placement in flash memory. Neuralink leverages the concept of Neuron Co-Activation, where neurons frequently activated together are linked to facilitate continuous read access and optimize I/O efficiency. Our approach incorporates a two-stage solution: an offline stage that reorganizes neuron placement based on co-activation patterns, and an online stage that employs tailored data access and caching strategies to align well with hardware characteristics. Evaluations conducted on a variety of smartphones and LLMs demonstrate that Neuralink achieves on average $1.49\times$ improvements in end-to-end latency compared to the state-of-the-art. As the first solution to optimize storage placement under sparsity, Neuralink explores a new optimization space at the intersection of sparsity-driven algorithm and storage-level system co-design for LLM inference.

cs.LG↗

arXiv2025

Video-in-the-Loop: Span-Grounded Long Video QA with Interleaved Reasoning

We present \emph{Video-in-the-Loop} (ViTL), a two-stage long-video QA framework that preserves a fixed token budget by first \emph{localizing} question-relevant interval(s) with a low-fps skim and then \emph{answering} via span-aware reallocation of visual tokens at higher effective frame rate, emitting an interleaved output with both spans and the final option for direct attribution. We also introduce \dataname{}, which converts description based event graphs into \emph{span-grounded} multiple-choice QA by pairing each question with \emph{ground-truth} time span(s) and related reasoning. ViTL is trained end-to-end with an interleaved group-relative objective that couples temporal IoU for localization with answer correctness, allowing credit to flow from answers back to spans without increasing compute. Under fixed token budgets, ViTL attains up to 8.6% with 50% less frame input on long-video QA and temporal grounding (e.g., Charades-STA, ActivityNet-Captions) and ablations show that span-aware token reallocation consistently surpasses uniform sampling. Together, \dataname{} and ViTL provide an interpretable, compute-efficient recipe for scalable long-video QA.

cs.CV↗

arXiv2025

AdaNav: Adaptive Reasoning with Uncertainty for Vision-Language Navigation

Vision Language Navigation (VLN) requires agents to follow natural language instructions by grounding them in sequential visual observations over long horizons. Explicit reasoning could enhance temporal consistency and perception action alignment, but reasoning at fixed steps often leads to suboptimal performance and unnecessary computation. To address this, we propose AdaNav, an uncertainty-based adaptive reasoning framework for VLN. At its core is the Uncertainty Adaptive Reasoning Block (UAR), a lightweight plugin that dynamically triggers reasoning. We introduce Action Entropy as a policy prior for UAR and progressively refine it through a Heuristics to RL training method, enabling agents to learn difficulty aware reasoning policies under the strict data limitations of embodied tasks. Results show that with only 6K training samples, AdaNav achieves substantial gains over closed source models trained on million scale data, improving success rate by 20% on R2R val-unseen, 11.7% on RxR-CE, and 11.4% in real world scenes. The code is available at https://github.com/xinding-sys/AdaNav.

cs.RO↗

arXiv2025

Sculptor: Empowering LLMs with Cognitive Agency via Active Context Management

Large Language Models (LLMs) suffer from significant performance degradation when processing long contexts due to proactive interference, where irrelevant information in earlier parts of the context disrupts reasoning and memory recall. While most research focuses on external memory systems to augment LLMs' capabilities, we propose a complementary approach: empowering LLMs with Active Context Management (ACM) tools to actively sculpt their internal working memory. We introduce Sculptor, a framework that equips LLMs with three categories of tools: (1) context fragmentation, (2) summary, hide, and restore, and (3) precise search. Our approach enables LLMs to proactively manage their attention and working memory, analogous to how humans selectively focus on relevant information while filtering out distractions. Experimental evaluation on diverse long-context benchmarks demonstrates that Sculptor significantly improves performance even without specific training, leveraging LLMs' inherent tool-calling and instruction-following capabilities. To further optimize these strategies, we introduce a novel dynamic context-aware reinforcement learning (RL) approach, advancing the training of an agent that actively modifies its own conversational history. By enabling Active Context Management, Sculptor not only mitigates proactive interference but also provides a cognitive foundation for more reliable reasoning across diverse long-context tasks-highlighting that explicit context-control strategies, rather than merely larger token windows, are key to robustness at scale.

cs.CL↗

arXiv2025

Scaling LLM Test-Time Compute with Mobile NPU on Smartphones

Deploying Large Language Models (LLMs) on mobile devices faces the challenge of insufficient performance in smaller models and excessive resource consumption in larger ones. This paper highlights that mobile Neural Processing Units (NPUs) have underutilized computational resources, particularly their matrix multiplication units, during typical LLM inference. To leverage this wasted compute capacity, we propose applying parallel test-time scaling techniques on mobile NPUs to enhance the performance of smaller LLMs. However, this approach confronts inherent NPU challenges, including inadequate hardware support for fine-grained quantization and low efficiency in general-purpose computations. To overcome these, we introduce two key techniques: a hardware-aware tile quantization scheme that aligns group quantization with NPU memory access patterns, and efficient LUT-based replacements for complex operations such as Softmax and dequantization. We design and implement an end-to-end inference system that leverages the NPU's compute capability to support test-time scaling on Qualcomm Snapdragon platforms. Experiments show our approach brings significant speedups: up to 19.0 for mixed-precision GEMM and 2.2 for Softmax. More importantly, we demonstrate that smaller models using test-time scaling can match or exceed the accuracy of larger models, achieving a new performance-cost Pareto frontier.

cs.DC↗