arXiv subjects

Xiang Li

Publications and source records attributed to Xiang Li.

At least 73 records · Page 4Linked to original sources

arXiv2026

SOLAR: SVD-Optimized Lifelong Attention for Recommendation

Attention mechanism remains the defining operator in Transformers since it provides expressive global credit assignment, yet its quadratic cost in sequence length N makes long-context modeling expensive and often forces truncation or other heuristics. Linear attention reduces complexity to O(Nd^2) by reordering computation through kernel feature maps, but this reformulation drops the softmax mechanism and shifts the attention score distribution. Lifelong recommendation requires efficient attention as well, for large-scale sequence modeling with user histories and candidate items under tight latency and resource constraints. We introduce SVD-Attention, a novel attention mechanism, and SOLAR, a set-aware framework built on it for lifelong recommendation. SVD-Attention factorizes low-rank embeddings into r principal components, computes candidate-to-interest scores in compact space, and applies softmax over those scores. Its bilinear reduction is exact on the rank-r reconstruction, while the normalized output approximates token-level softmax with an explicitly bounded residual. The resulting computation reduces the cost from O(N^2d) to O(Ndr), and supports 12,000 behaviors and 3,000 candidates per request without filtering. SOLAR achieves the best among compared methods on RecFlow and MIND, delivers 0.8531 AUC at approximately 19ms 95th-percentile latency in an industrial evaluation, and yields business gains with a 0.68% relative lift in Video Views in the real-world online A/B test. Following this evaluation, SOLAR has been fully deployed in Kuaishou's production recommendation system.

cs.IR↗

arXiv2026

TensorHub: Scalable and Elastic Weight Transfer for LLM RL Training

Modern LLM reinforcement learning (RL) workloads require a high-performance weight transfer system to scale training across heterogeneous compute resources. However, efficiently transferring terabyte-scale model weights across thousands of GPUs remains challenging because the system must accommodate clusters that dynamically scale up and down while keeping coordination, data movement, and storage overhead low. We introduce Reference-Oriented Storage (ROS), a new storage abstraction for RL weight transfer that exploits highly replicated model weights in place. ROS presents the illusion that certain versions of the model weights are stored and can be fetched on demand. Underneath, ROS does not physically store any copies of the weights; instead, it tracks the workers that hold these weights on GPUs for inference. Upon request, ROS directly uses them to serve reads. We build TensorHub, a production-quality system that instantiates the ROS idea with topology-aware transfer, model-parallel consistency, and fault tolerance. Evaluation shows that TensorHub saturates RDMA bandwidth and adapts to three distinct rollout workloads with minimal engineering effort. Specifically, TensorHub reduces total GPU stall time by up to 6.7x for standalone rollouts, accelerates weight updates for elastic rollouts by up to 4.8x, and cuts cross-datacenter rollout stall time by up to 19x. TensorHub has been deployed in ByteDance production to support cutting-edge RL training.

cs.DC↗

arXiv2026

Relation-Aware Graph Foundation Model

In recent years, large language models (LLMs) have demonstrated remarkable capability to generalize across diverse natural language processing tasks, inspiring the development of graph foundation models (GFMs) for large-scale pre-training. However, unlike language models with explicit token units, graphs lack a well-defined unit for generalization, making it challenging to design effective pre-training strategies. In this work, we propose REEF, a novel GFM framework that leverages relation tokens as the fundamental units. We construct a vocabulary of relation tokens to encode relational information within graphs. To accommodate diverse relations, we introduce two hypernetworks that adaptively generate the parameters of aggregators and classifiers in graph neural networks based on relation tokens. In addition, we design another hypernetwork to construct dataset-specific projectors and incorporate a dataset-level feature bias into the initial node representations, enhancing flexibility across different datasets with the same relation. Extensive experiments demonstrate that REEF consistently outperforms existing methods in both pre-training and transfer learning, highlighting its potential as a general-purpose graph foundation model.

cs.LG↗

arXiv2026

Dynamic Kuramoto-Hodge Operators for PDEs on Complex Geometries and Topologies

Learning PDE operators on complex domains requires capturing interactions among fields on vertices, edges, and faces, alongside global responses shaped by topology. Existing neural operators accommodate irregular geometries but often overlook these distinct field supports or their condition-dependent coupling. We introduce the Dynamic Kuramoto--Hodge Operator (DKHO), which combines topology-constrained interactions with learned coordination. DKHO encodes conditions on their native cochain supports, evolves Kuramoto-inspired relation states through the boundary and coboundary operators that compose the Dirac operator, and decodes non-harmonic and harmonic responses in orthogonal Hodge subspaces. Topology thus determines where information can flow, while learned dynamics adapts how it is exchanged to each PDE instance. Across porous-medium Darcy flow, torus transport--diffusion, and cavity magnetostatics, DKHO-large reduces prediction error by approximately 61% on average over leading baselines, while DKHO-small remains competitive using only 11.5--24.3% as many parameters. These results suggest that coupling topological structure with adaptive dynamics provides an effective inductive bias for accurate and parameter-efficient PDE operator learning on complex geometries and topologies.

cs.LG↗

arXiv2026

How code helps different tasks? A decompositional lens on LLM post-training

Evaluating code data as a single corpus can obscure which types of code data benefit which models and downstream tasks. Effective data selection requires understanding both the benefits of individual categories and whether these benefits persist when categories are combined. We introduce a decompositional lens for studying these effects in LLM post-training. We first decompose an execution-verified code corpus into interpretable categories based on the computational patterns of its solutions. Through controlled fine-tuning experiments, we compare individual categories with a balanced mixture across instruction-tuned models on question answering, mathematics, and code generation. The resulting response maps reveal recurring gains in average question-answering performance, while the same category can improve one model or task and degrade another. The best-performing category also varies with the starting model and target task. We then compose compact mixtures guided by these results and examine whether benefits observed in individual categories persist under joint training. On selected model--task pairs, mixtures whose constituents each improve the target task outperform both their best constituent and full-corpus training while using roughly 10--15\% of the full corpus. These exploratory findings illustrate a \emph{less is more} pattern and highlight how the value of code data in post training depends on which categories are combined for which model and task.

cs.AI↗

arXiv2026

HiMed: Incentivizing Hindi Reasoning in Medical LLMs

Medical large language models hold promise for reducing healthcare disparities, yet Hindi remains severely underrepresented. While medical LLMs excel in high-resource languages, their performance degrades sharply in Hindi, particularly on Indian systems of medicine. We argue that robust cross-lingual medical transfer requires Hindi reasoning. To this end, we introduce HiMed, a Hindi reasoning medical corpus and benchmark suite covering both Western and Indian medicine. We further propose HiMed-8B, a Hindi-form medical reasoning LLM, through the design of decaying scaffolding reward. Extensive experiments demonstrate improvement in Hindi medical reasoning performance and a reduction in the English-Hindi accuracy gap. Ablation studies validate the contribution of each training stage and reward component. All code, data, and weights are available

cs.CL↗

arXiv2026

Not Every Token Is Worth Distilling: Selective Supervision for Direct-OPD

Direct On-Policy Distillation (Direct-OPD) transfers reinforcement-learning-induced policy improvements from a small model to a larger student by using the token-level log-ratio between post-RL and pre-RL checkpoints as dense supervision on the student's own rollouts. This transfer rewards the policy shift at every state, yet the log-ratio measures only relative change: it can stay fixed even as the probability mass that both checkpoints assign to the student's candidate tokens vanishes. Through an exact construction, we show that the Direct-OPD reward and its update can remain unchanged while the Jensen-Shannon divergence (JSD) and both KL directions between the checkpoints vanish with this mass, and we note that a small JSD bounds how much the teacher's behavior changed. Motivated by this analysis, we propose Selective Supervision for Direct-OPD (S$^2$D-OPD), which ranks student-sampled states by their teacher-reference JSD and masks Direct-OPD supervision at low-divergence states, retaining only the top 10% of states per response. Across two teacher pairs and four student models ranging from 1.7B to 8B parameters, S$^2$D-OPD improves held-out accuracy over dense Direct-OPD on AIME and HMMT benchmarks in seven of eight settings and matches it in the eighth, without extra forward passes.

cs.LG↗

arXiv2026

FoundDSR: A Generalizable Foundation Model with Guided 2D Gaussian Splatting for Depth Super-Resolution

We introduce FoundDSR, a generalizable foundation model for robust depth reconstruction across unseen data distributions using RGB-D pairs. FoundDSR begins with a guided 2D Gaussian Splatting strategy to model depth representations with Gaussian primitives. This strategy employs high-resolution RGB as prompts to optimize the Gaussian parameters, thereby encouraging each Gaussian primitive to anisotropically deform along high-frequency structural directions. The resulting Gaussian-upsampled representations are then mapped to high-resolution depth through an effective depth reconstruction branch. Furthermore, to mitigate training instability and bias toward dominant sources caused by distribution gaps in large-scale heterogeneous data, we introduce heterogeneous federated learning that allocates each data source to an independent client for local optimization and global aggregation. This design effectively endows FoundDSR with stable scalability to diverse and large-scale training data. Extensive zero-shot evaluations on synthetic, real-world, arbitrary-scale, and noisy conditions demonstrate that FoundDSR consistently outperforms existing state-of-the-art approaches, confirming its strong robustness and generalization to unknown scenes.

cs.CV↗

arXiv2026

CAIRN: Dynamic Fact-Intent DAGs for Multi-Agent Exploration

LLM-powered autonomous systems have demonstrated promising capabilities in mathematical reasoning, engineering, and cybersecurity. Yet how to organize these systems for effective, reliable, and sustained performance remains an open question. In this paper, we present CAIRN, a fact-intent-driven multi-agent paradigm for goal-directed exploration. CAIRN represents observations and planned investigations as a dynamic directed acyclic graph (DAG). A reasoner interprets facts to propose intents, which workers execute to produce new facts. Each intent references its supporting facts and defines a potential exploration branch. The persistent graph preserves goals, dependencies and findings across workers, supporting knowledge reuse and parallel exploration. The graph also makes execution trajectories traceable and auditable, providing a basis for human verification and intervention. We evaluate CAIRN across cybersecurity and mathematical reasoning tasks, examining task success, time to solution, and token consumption. DAG-based coordination can incur higher token costs with no observable performance gains on tasks that require little effort. However, on high-effort tasks (at least 1M tokens), we observe faster solutions in 76.5% of cases, with speedups of up to 3.08x. Moreover, as task effort increases, these time gains become more pronounced while relative token overhead declines, highlighting the potential of DAG-guided parallel exploration.

cs.AI↗

arXiv2026

Q-Probe: Scaling Image Quality Assessment to High Resolution via Context-Aware Agentic Probing

Reinforcement Learning (RL) has empowered Multimodal Large Language Models (MLLMs) to achieve superior human preference alignment in Image Quality Assessment (IQA). However, existing RL-based IQA models typically rely on coarse-grained global views, failing to capture subtle local degradations in high-resolution scenarios. While emerging "Thinking with Images" paradigms enable multi-scale visual perception via zoom-in mechanisms, their direct adaptation to IQA induces spurious "cropping-implies-degradation" biases and misinterprets natural depth-of-field as artifacts. To address these challenges, we propose Q-Probe, the first agentic IQA framework designed to scale IQA to high resolution via context-aware probing. First, we construct Vista-Bench, a pioneering benchmark tailored for fine-grained local degradation analysis in high-resolution IQA settings. Furthermore, we propose a three-stage training paradigm that progressively aligns the model with human preferences, while simultaneously eliminating causal bias through a novel context-aware cropping strategy. Extensive experiments demonstrate that Q-Probe achieves state-of-the-art performance in high-resolution settings while maintaining superior efficacy across resolution scales.

eess.IV↗

arXiv2026

BioEVAL: A global, multi-institutional benchmark of large language and multimodal models for bioengineering

Large Language Models (LLMs) have demonstrated historic breakthroughs in general reasoning with early successes in biomedical science. However, existing LLM benchmarking emphasizes factual recall, offering limited insight into model performance on frontier and multimodal tasks. We assembled BioEVAL (BioEngineering Validation of AI and LLMs), a global, multi-institutional initiative designed to assess experimental reasoning capability across bioengineering (BE) subfields. BioEVAL spans 11 major BE subfields plus a set of uncategorized items, bringing together 22 research groups to create a PhD-level benchmark comprising 608 evaluation items: 1) 380 multiple-choice questions (MCQs, 359 retained after audit), 2) 218 literature synthesis tasks, and 3) 10 multimodal problems with experimental image interpretation. Benchmark items underwent authoring-group expert review and centralized quality control before evaluation. Following evaluation, a blinded cross-group consensus audit of the highest- and lowest-accuracy MCQ items flagged 21 questions for revision or removal; these were withheld, and all reported MCQ results are computed on the 359 retained items. We evaluated diverse cloud-scale foundation/multimodal models (e.g., ChatGPT, Gemini, and Grok) and locally deployable models suitable for inference on consumer-grade GPUs. Models achieved the highest accuracy of up to 90% on MCQs, similarity score of 0.72 on literature synthesis, and accuracy of 80% on a small sample of multimodal reasoning questions, with substantial performance variation across subfields. Leaderboard rankings characterize current capabilities, limitations, and development priorities across the evaluated BE task categories. BioEVAL is maintained as an extensible benchmark with standardized protocols for continuing expert item contribution and model evaluation.

cs.AI↗

arXiv2026

Evaluation of OpenAI o1: Opportunities and Challenges of AGI

This comprehensive study evaluates the performance of OpenAI's o1-preview large language model across a diverse array of complex reasoning tasks, spanning multiple domains, including computer science, mathematics, natural sciences, medicine, linguistics, and social sciences. Through rigorous testing, o1-preview demonstrated remarkable capabilities, often achieving human-level or superior performance in areas ranging from coding challenges to scientific reasoning and from language processing to creative problem-solving. Key findings include: -83.3% success rate in solving complex competitive programming problems, surpassing many human experts. -Superior ability in generating coherent and accurate radiology reports, outperforming other evaluated models. -100% accuracy in high school-level mathematical reasoning tasks, providing detailed step-by-step solutions. -Advanced natural language inference capabilities across general and specialized domains like medicine. -Impressive performance in chip design tasks, outperforming specialized models in areas such as EDA script generation and bug analysis. -Remarkable proficiency in anthropology and geology, demonstrating deep understanding and reasoning in these specialized fields. -Strong capabilities in quantitative investing. O1 has comprehensive financial knowledge and statistical modeling skills. -Effective performance in social media analysis, including sentiment analysis and emotion recognition. The model excelled particularly in tasks requiring intricate reasoning and knowledge integration across various fields. While some limitations were observed, including occasional errors on simpler problems and challenges with certain highly specialized concepts, the overall results indicate significant progress towards artificial general intelligence.

cs.CL↗

arXiv2026

M3D-Net: Hierarchical Coordination of Spatial Context, Feature Reuse, and Differential Attention for Mammography Classification

Breast image classification requires local detail and global tissue context, yet these cues can weaken as representations deepen. We present M3D-Net, a mammography encoder that hierarchically coordinates multi-scale coordinate attention, bounded dynamic feature reuse, and differential attention through resolution-aware operator placement. Within-stage retrieval preserves access to earlier features, coordinate-aware aggregation integrates local and global context, and differential attention operates at coarse resolutions. We evaluate image-only classification on AISSLab mammography and an adapted image--clinical model on BrEaST ultrasound. Against EdgeNeXt, RepViT, and TransXNet, the proposed implementations achieve the highest recorded validation accuracy and late-training accuracy, with the lowest endpoint cross-entropy loss. Validation accuracies reach 97.78\% and 80.39\%, respectively. These results support further evaluation of hierarchical coordination across breast imaging settings; repeated-seed, component-controlled, and independent evaluations remain necessary.

cs.CV↗

arXiv2026

Order-Dependent Modification of High-Harmonic Generation by Quantum Dissipation and Lamb Shift

Many-body environments are conventionally incorporated into quantum dynamics as heat baths, which induce a Lamb shift and quantum dissipation. In the traditional picture, such environmental coupling is expected to induce suppression and broadening of spectral signals. In this letter, we investigate high-order harmonic generation (HHG) in a two-level system coupled to a heat bath via dipole-dipole interactions using the Lindblad master equation. It is found that the environmental effect does not simply suppress the harmonic efficiency. Instead, when the cutoff frequency of the coupling spectral density exceeds the energy-level spacing, the environment can actually enhance the harmonic yield. Further analysis reveals that this enhancement originates from intense level fluctuations induced by the Lamb shift. Meanwhile, the environmental influence exhibits a distinct dependence on harmonic order, manifesting clearly different behaviors for lower and higher harmonics. Our results challenge the common relaxation-time picture of uniform damping and establish a microscopic, order-dependent mechanism for environmental control of HHG, with direct implications for solid-state attosecond spectroscopy and quantum material engineering.

physics.optics↗

arXiv2026

Nature of small-scale Ca II H brightenings associated with weak fields in a quiet Sun region

Context. Chromospheric emission observed in the Ca II H and K resonance lines is generally positively correlated with photospheric magnetic fields. However, a fraction of strong Ca II H&K brightenings are found to occur over regions of low photospheric magnetic flux density. Aims. We aim to understand the origin of these chromospheric brightenings associated with weak fields in quiet Sun regions. Methods. We use Ca II H observations taken by the SuFI along with the photospheric magnetograms recorded by the IMaX, instruments onboard the first flight of the balloon-borne observatory Sunrise. We also use the atmospheric physical parameters inverted from IMaX data by the SPINOR code. Results. 217 Ca II H brightenings associated with weak fields are identified. 111 of these are located at the peripheries of strong field regions, in the magnetic canopies formed by the expansion of the kG flux tube. The other brightenings are associated with either weak unipolar magnetic patches, with interactions between weak mixed-polarity magnetic fields, or at locations with no clear magnetic features. For the latter two situations, the Ca II H intensity varies with a period of 3-5 mins, accompanied by variations in the line-of-sight velocity, the temperature, and the magnetic flux with a similar period. Conclusions. About half of the strong Ca II H brightenings associated with weak fields are located in magnetic canopies overlying very weak photospheric fields. About 10-12% are associated with magnetic bright points where the magnetic field is probably not spatially resolved. The remainder appears to be Ca II H grains, which are overwhelmingly associated with magnetic emergence or cancellation events. We conclude that although Ca II grains are mainly driven by (magneto-)acoustic shock waves, interactions between opposite-polarity magnetic fields also contribute to the formation of the brightest peaks.

astro-ph.SR↗

arXiv2026

Speculative Evaluation of Stochastic LLMs

Evaluating a stochastic large language model is costly: benchmark scores estimate expected performance from randomized rollouts, yet uniform repetition ignores sharp differences in task-level rollout variance. We ask how to minimize the variance of a fixed-benchmark mean under an exact rollout budget. We develop Speculative Evaluation with a Hierarchical Bayesian Neyman (HBN) policy with pilot size and stage weight jointly chosen ex ante. It runs a short uniform pilot, pools per-task success counts with a hierarchical Bayesian model, and uses posterior expectations of task-level sampling variances for exact positive-integer Neyman allocation. To mitigate the pilot synchronization barrier, HBN-async speculatively executes continuations from partial pilot feedback and retains those selected by the final allocation. Across six checkpoints and 18 benchmark groups, we evaluate 107 nondegenerate benchmark-checkpoint profiles. For rollout budgets of 8-64 per task, Speculative Evaluation reduces variance relative to Uniform by 12.8%-33.6% on average across profiles, outperforming hindsight-tuned empirical and independent Bayesian baselines. Real-generation experiments that account for the pilot synchronization barrier show that HBN-async mitigates its overhead, helping translate statistical efficiency into practical evaluation benefits.

stat.ML↗

arXiv2026

StepKV: Step-Aware KV Cache Compression for LLM Agents

Key-value (KV) caching is essential for efficient autoregressive large language model (LLM) inference, but the cache grows linearly with context length, increasing storage and decoding costs. KV cache compression mitigates this cost by retaining only a subset of cached tokens. This challenge is particularly important for multi-step LLM agents, where a query expands into trajectories of reasoning, tool interactions, and retrieved observations. Existing pruning methods typically treat the cache as a flat token stream and rank tokens by recency or attention saliency. This creates a mismatch between the unit of compression and the unit of reasoning: token-level pruning removes individual entries, whereas useful information in multi-step agents is often organized into reasoning steps with uneven and delayed importance. Consequently, an early observation or intermediate decision may receive little recent attention yet remain essential for later evidence synthesis. We term this failure mode Reasoning Continuity Disruption.These observations motivate KV cache compression that jointly considers token- and reasoning-step-level information. StepKV addresses this goal by treating reasoning steps as first-class retention units. It associates cache entries with their generating steps, estimates step utility from trajectory-derived signals, and combines this utility with token-level saliency. The resulting scores globally rank prunable tokens, from which StepKV retains the top-scoring entries under a target budget. StepKV thus provides a step-centric perspective for agent KV cache compression. Across multi-hop QA and long-horizon web reasoning tasks, StepKV sustains accuracy under low KV budgets where token-level baselines degrade sharply, offering a more robust efficiency-accuracy trade-off for multi-step agent inference.

cs.LG↗

arXiv2026

Spoken Language Models that Think Aloud

While Chain-of-Thought (CoT) reasoning has improved the capability of language models, directly applying it to Spoken Language Models (SLMs) may introduce long silent intervals under the serial "think-then-speak" paradigm, disrupting real-time spoken interaction. To address this issue, we propose an asynchronous think-aloud framework for reasoning-based SLMs within the Thinker-Talker architecture. The framework maintains a primary reasoning stream for logical deduction and a lightweight think-aloud stream that generates short, task-grounded progress utterances conditioned on the user input and the evolving reasoning state. A dynamic balance strategy coordinates the two streams at runtime, triggering additional think-aloud speech to avoid silent gaps and canceling pending utterances when the final response becomes ready. Experiments on spoken reasoning and question-answering benchmarks show that our approach substantially reduces user-audible silence during reasoning while maintaining answer accuracy comparable to that of a serial "think-then-speak" baseline, demonstrating the potential of asynchronous think-aloud for responsive interaction in SLMs.

cs.CL↗