arXiv subjects

Xiaoran Liu

Publications and source records attributed to Xiaoran Liu.

At least 19 recordsLinked to original sources

arXiv2026

Faster Than Flash: Exploiting Attention Sparsity for Efficient Long-Context Decoding

The development of long-context Large Language Models (LLMs) is constrained by the memory bandwidth bottleneck and quadratic complexity of the attention mechanism during decoding. To overcome the inherent trade-offs between the memory overhead of metadata-based metrics and the computational inefficiency of adaptive selection strategies, we present Faster Flash Decoding (FFD), a novel hardware-algorithm co-design framework designed to break the memory wall in long-context decoding. FFD integrates the selector and computer into a fully fused kernel, replacing external metadata indices with content-aware scanning via low-bit quantization. Furthermore, we introduce the top-delta strategy, which dynamically filters blocks to achieve distribution-adaptive sparsity without global synchronization. Offering a training-free and plug-and-play solution, FFD also enables the reuse of scanning results for computation, achieving up to 11.6x kernel-level speedup and scaling to 256K context length, with 2.37x end-to-end throughput improvement. Empirical validation on RULER and LongBench confirms that FFD maintains model accuracy while delivering high-ratio sparsity, with code available at https://github.com/qluoluo/faster-flash-decoding

cs.LG↗

arXiv2026

FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation

Despite the non-autoregressive potential of diffusion language models (dLLMs), existing decoding strategies demonstrate positional bias, failing to fully unlock the potential of arbitrary generation. In this work, we delve into the inherent spectral characteristics of dLLMs and present the first frequency-domain analysis showing that low-frequency components in hidden states primarily encode global structural information and long-range dependencies, while high-frequency components are responsible for characterizing local details. Based on this observation, we propose FourierSampler, which leverages a frequency-domain sliding window mechanism to dynamically guide the model to achieve a "structure-to-detail" generation. FourierSampler outperforms other inference enhancement strategies on LLADA and SDAR, achieving relative improvements of 20.4% on LLaDA1.5-8B and 16.0% on LLaDA-8B-Instruct. It notably surpasses similarly sized autoregressive models like Llama3.1-8B-Instruct.

cs.CL↗

arXiv2026

AI Can Learn Scientific Taste

Scientific discovery depends on expert judgement and foresight, which we call scientific taste: the ability to judge and propose research ideas with the potential for long-term scientific impact. Scientific taste is largely concentrated among highly experienced researchers, whose expertise is usually limited to a few specialised fields. If AI could learn scientific taste, it could reduce reliance on human experts and accelerate scientific discovery. Whether AI can learn this ability remains an open question. We introduce Reinforcement Learning from Community Feedback (RLCF) to learn judgement and ideation. Scientific Judge learns from community feedback, such as citations. Scientific Thinker learns to propose research ideas with high potential impact. Experiments show that Scientific Judge outperforms strong LLM baselines and that learned judgement generalises to future-year papers, other community metrics, and unseen fields. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than those proposed by baselines. These results suggest that AI can learn scientific taste, marking an important step towards AI systems that could help accelerate scientific discovery.

cs.CL↗

arXiv2026

Writing and erasing skyrmions by single ultrafast laser pulses in monolayer Janus 2D magnets

Skyrmions in 2D magnets are promising candidates for nonvolatile, low-power, and high-density spintronic memories. However, their experimental realization at the 2D limit remains challenging, owing to the difficulty in engineering the required chiral magnetic interactions. Here, we report the creation and direct imaging of Néel-type skyrmions in Janus 2D chromium chalcogenides using synchrotron X-ray photoemission electron microscopy, and scanning nitrogen-vacancy magnetometry, which exhibit field-free stability, nonvolatility, and size tunability. First-principles calculations and micromagnetic simulations reveal that Janus-surface-induced inversion-symmetry breaking enhances the Dzyaloshinskii-Moriya interaction, providing the microscopic mechanism for skyrmion stabilization and tunability. We further achieve reversible skyrmion writing and erasing using a single ultrafast laser pulse in a magnetic field as low as 300 Oe, demonstrating the excellent manipulability of this 2D magnetic system. These results establish Janus engineering as a route to creating and manipulating nonvolatile skyrmions in atomically thin magnets, with implications for skyrmion-based low-power spintronic devices.

cond-mat.mtrl-sci↗

arXiv2026

A Model-Driven Approach for Developing Families of Reinforcement Learning Environments

Virtual training environments are software-intensive systems in which reinforcement learning (RL) agents learn, adapt, and demonstrate meaningful behavior. Virtual training environments offer a safe and cost-efficient alternative to training agents in real-world settings. However, to converge, most realistic RL problems require training in multiple, mostly similar but slightly different environments - i.e., families of environment variants. The typical development process of environment families is a labor-intensive and error-prone manual endeavor that does not scale well. To alleviate these issues, in this paper, we propose a model-driven approach for developing families of RL training environments. To obtain the family of environments, we develop an approach and prototype tool. In our approach, a hybrid genetic algorithm - a combination of population-based global search and heuristic local search - generates environment families. Mutations and constraints are expressed as model transformations and are operationalized into a search process by a state-of-the-art model transformation engine. We demonstrate the soundness of our approach in a wildfire mitigation scenario and curriculum learning - a particular learning paradigm that relies on environment families.

cs.SE↗

arXiv2026

Designing Quantum Matter in Pyrochlore Iridates: A Perspective on Recent Thin-Film Advances

The pyrochlore iridates R2Ir2O7 have emerged as a unique playground for exploring exotic quantum phenomena arising from the intricate interplay of strong spin-orbit coupling, electron correlations, and geometric frustration. While bulk crystals of these materials have revealed a rich landscape of correlated and topological states, recent breakthroughs in epitaxial thin-film synthesis and heterostructure engineering unlocked an entirely new dimension of discovery. This brief Perspective reviews recent advancements highlighting how new tuning knobs such as dimensional confinement, epitaxial strain, and interfacial coupling can be used to manipulate the delicate balance of competing interactions. We discuss several key discoveries enabled by this approach including the realization of the magnetic Weyl semimetal phase in (111) oriented films, strain-engineered magnetic multipolar orders, the emergence of a chiral spin liquid-like state in the quasi-2D limit, and the discovery of novel electronically anisotropic states at interfaces between pyrochlore iridates and other quantum materials, such as spin ice pyrochlores. These findings showcase that low-dimensional pyrochlore iridates provide ample opportunities for both theory and experiment to unravel, control and ultimately design novel quantum states of matter. We conclude by outlining key open questions and future directions ranging from the synthesis of new heterostructures to the application of advanced probes and the exploration of non-equilibrium phenomena.

cond-mat.str-el↗

arXiv2026

Prism: Spectral-Aware Block-Sparse Attention

Block-sparse attention is promising for accelerating long-context LLM pre-filling, yet identifying relevant blocks efficiently remains a bottleneck. Existing methods typically employ coarse-grained attention as a proxy for block importance estimation, but often resort to expensive token-level searching or scoring, resulting in significant selection overhead. In this work, we trace the inaccuracy of standard coarse-grained attention via mean pooling to a theoretical root cause: the interaction between mean pooling and Rotary Positional Embeddings (RoPE). We prove that mean pooling acts as a low-pass filter that induces destructive interference in high-frequency dimensions, effectively creating a "blind spot" for local positional information (e.g., slash patterns). To address this, we introduce Prism, a training-free spectral-aware approach that decomposes block selection into high-frequency and low-frequency branches. By applying energy-based temperature calibration, Prism restores the attenuated positional signals directly from pooled representations, enabling block importance estimation using purely block-level operations, thereby improving efficiency. Extensive evaluations confirm that Prism maintains accuracy parity with full attention while delivering up to $\mathbf{5.1\times}$ speedup.

cs.CL↗

arXiv2026

Discovery of unobservable parameters via physical embedding

Recovering a source signal from indirect measurements often requires estimating latent parameters, such as wireless channel states or MRI coil sensitivities, that cannot be directly observed. Here, we introduce Physics-Embedded Inverse Learning (PEIL), in which a learned estimator predicts these parameters and a fixed, physics-based inverse operator uses them to reconstruct the signal, so that training requires only the source signal as supervision. In systems where multiple parameter combinations can reconstruct the signal equally well, the estimator exploits this freedom to coordinate parameters that compensate for residual modelling errors rather than match ground-truth parameters. In high-mobility wireless communications, PEIL discovers task-optimal configurations that outperform baselines given access to ground-truth parameters, enabling zero-shot generalisation and over 20-fold reduction in training data relative to supervised baselines. To test whether these properties extend across physical domains, we apply PEIL to parallel MRI, where it discovers physically interpretable coil sensitivity maps without calibration scans, yielding reconstructions grounded purely in acquired measurements. These results demonstrate that non-identifiability, conventionally a liability, becomes a resource when the learning objective targets reconstruction quality rather than parameter accuracy.

eess.SP↗

arXiv2026

Trust the AI, Doubt Yourself: The Effect of Urgency on Self-Confidence in Human-AI Interaction

Studies show that interactions with an AI system fosters trust in human users towards AI. An often overlooked element of such interaction dynamics is the (sense of) urgency when the human user is prompted by an AI agent, e.g., for advice or guidance. In this paper, we show that although the presence of urgency in human-AI interactions does not affect the trust in AI, it may be detrimental to the human user's self-confidence and self-efficacy. In the long run, the loss of confidence may lead to performance loss, suboptimal decisions, human errors, and ultimately, unsustainable AI systems. Our evidence comes from an experiment with 30 human participants. Our results indicate that users may feel more confident in their work when they are eased into the human-AI setup rather than exposed to it without preparation. We elaborate on the implications of this finding for software engineers and decision-makers.

cs.AI↗

arXiv2026

A Reference Architecture of Reinforcement Learning Frameworks

The surge in reinforcement learning (RL) applications gave rise to diverse supporting technology, such as RL frameworks. However, the architectural patterns of these frameworks are inconsistent across implementations and there exists no reference architecture (RA) to form a common basis of comparison, evaluation, and integration. To address this gap, we propose an RA of RL frameworks. Through a grounded theory approach, we analyze 18 state-of-the-practice RL frameworks and, by that, we identify recurring architectural components and their relationships, and codify them in an RA. To demonstrate our RA, we reconstruct characteristic RL patterns. Finally, we identify architectural trends, e.g., commonly used components, and outline paths to improving RL frameworks.

cs.SE↗

arXiv2026

Developing AI Agents with Simulated Data: Why, what, and how?

As insufficient data volume and quality remain the key impediments to the adoption of modern subsymbolic AI, techniques of synthetic data generation are in high demand. Simulation offers an apt, systematic approach to generating diverse synthetic data. This chapter introduces the reader to the key concepts, benefits, and challenges of simulation-based synthetic data generation for AI training purposes, and to a reference framework to describe, design, and analyze digital twin-based AI simulation solutions.

cs.AI↗

arXiv2026

Introduction to Digital Twins for the Smart Grid

This chapter provides an introduction to the foundations of digital twins and makes the case for employing them in smart grids. As engineered systems become more complex and autonomous, digital twin technology gains importance as the unified technological platform for design, testing, operation, and maintenance. Smart grids are prime examples of such complex systems, in which unique design and operation challenges arise from the combination of physical and software components. As high-fidelity in-silico replicas of physical components, digital twins provide safe and cost-efficient experimentation facilities in the design and verification phase of smart grids. In the operation phase of smart grids, digital twins enable automated load balancing of grids through real-time simulation and decision-making. These, and an array of similar benefits, position digital twins as crucial technological components in smart grids.

cs.ET↗

arXiv2026

AgentLongBench: A Controllable Long Benchmark For Long-Contexts Agents via Environment Rollouts

The evolution of Large Language Models (LLMs) into autonomous agents necessitates the management of extensive, dynamic contexts. Current benchmarks, however, remain largely static, relying on passive retrieval tasks that fail to simulate the complexities of agent-environment interaction, such as non-linear reasoning and iterative feedback. To address this, we introduce \textbf{AgentLongBench}, which evaluates agents through simulated environment rollouts based on Lateral Thinking Puzzles. This framework generates rigorous interaction trajectories across knowledge-intensive and knowledge-free scenarios. Experiments with state-of-the-art models and memory systems (32K to 4M tokens) expose a critical weakness: while adept at static retrieval, agents struggle with the dynamic information synthesis essential for workflows. Our analysis indicates that this degradation is driven by the minimum number of tokens required to resolve a query. This factor explains why the high information density inherent in massive tool responses poses a significantly greater challenge than the memory fragmentation typical of long-turn dialogues.

cs.CL↗

arXiv2026

MrRoPE: Mixed-radix Rotary Position Embedding

Rotary Position Embedding (RoPE)-extension refers to modifying or generalizing the Rotary Position Embedding scheme to handle longer sequences than those encountered during pre-training. However, current extension strategies are highly diverse and lack a unified theoretical foundation. In this paper, we propose MrRoPE (Mixed-radix RoPE), a generalized encoding formulation based on a radix system conversion perspective, which elegantly unifies various RoPE-extension approaches as distinct radix conversion strategies. Based on this theory, we introduce two training-free extensions, MrRoPE-Uni and MrRoPE-Pro, which leverage uniform and progressive radix conversion strategies, respectively, to achieve 'train short, test long' generalization. Without fine-tuning, MrRoPE-Pro sustains over 85% recall in the 128K-context Needle-in-a-Haystack test and achieves more than double YaRN's accuracy on Infinite-Bench retrieval and dialogue subsets. Theoretical analysis confirms that MrRoPE-Pro effectively raises the upper bound of RoPE's attainable encoding length, which further validates the reliability and utility of our theory and methodology.

cs.CL↗

arXiv2026

DiRL: An Efficient Post-Training Framework for Diffusion Language Models

Diffusion Language Models (dLLMs) have emerged as promising alternatives to Auto-Regressive (AR) models. While recent efforts have validated their pre-training potential and accelerated inference speeds, the post-training landscape for dLLMs remains underdeveloped. Existing methods suffer from computational inefficiency and objective mismatches between training and inference, severely limiting performance on complex reasoning tasks such as mathematics. To address this, we introduce DiRL, an efficient post-training framework that tightly integrates FlexAttention-accelerated blockwise training with LMDeploy-optimized inference. This architecture enables a streamlined online model update loop, facilitating efficient two-stage post-training (Supervised Fine-Tuning followed by Reinforcement Learning). Building on this framework, we propose DiPO, the first unbiased Group Relative Policy Optimization (GRPO) implementation tailored for dLLMs. We validate our approach by training DiRL-8B-Instruct on high-quality math data. Our model achieves state-of-the-art math performance among dLLMs and surpasses comparable models in the Qwen2.5 series on several benchmarks.

cs.LG↗

arXiv2025

Beyond Real: Imaginary Extension of Rotary Position Embeddings for Long-Context LLMs

Rotary Position Embeddings (RoPE) have become a standard for encoding sequence order in Large Language Models (LLMs) by applying rotations to query and key vectors in the complex plane. Standard implementations, however, utilize only the real component of the complex-valued dot product for attention score calculation. This simplification discards the imaginary component, which contains valuable phase information, leading to a potential loss of relational details crucial for modeling long-context dependencies. In this paper, we propose an extension that re-incorporates this discarded imaginary component. Our method leverages the full complex-valued representation to create a dual-component attention score. We theoretically and empirically demonstrate that this approach enhances the modeling of long-context dependencies by preserving more positional information. Furthermore, evaluations on a suite of long-context language modeling benchmarks show that our method consistently improves performance over the standard RoPE, with the benefits becoming more significant as context length increases. The code is available at https://github.com/OpenMOSS/rope_pp.

cs.CL↗

arXiv2025

Thus Spake Long-Context Large Language Model

Long context is an important topic in Natural Language Processing (NLP), running through the development of NLP architectures, and offers immense opportunities for Large Language Models (LLMs), giving LLMs the lifelong learning potential akin to humans. Unfortunately, the pursuit of a long context is accompanied by numerous obstacles. Nevertheless, long context remains a core competitive advantage for LLMs. In the past two years, the context length of LLMs has achieved a breakthrough extension to millions of tokens. Moreover, research on long-context LLMs has expanded beyond length extrapolation to a comprehensive focus on architecture, infrastructure, training, and evaluation technologies. Inspired by the symphonic poem, Thus Spake Zarathustra, we draw an analogy between the journey of extending the context of LLM and the attempts of humans to transcend their mortality. In this survey, we will illustrate how LLM struggles between the tremendous need for a longer context and its equal need to accept the fact that it is ultimately finite. To achieve this, we give a global picture of the lifecycle of long-context LLMs from four perspectives: architecture, infrastructure, training, and evaluation, showcasing the full spectrum of long-context technologies. At the end of this survey, we will present 10 unanswered questions currently faced by long-context LLMs. We hope this survey can serve as a systematic introduction to research on long-context LLMs. Video: https://www.bilibili.com/video/BV11h9AYoEYj. Github: https://github.com/OpenMOSS/Thus-Spake-Long-Context-LLM.

cs.CL↗

arXiv2025

LongLLaDA: Unlocking Long Context Capabilities in Diffusion LLMs

Large Language Diffusion Models, or diffusion LLMs, have emerged as a significant focus in NLP research, with substantial effort directed toward understanding their scalability and downstream task performance. However, their long-context capabilities remain unexplored, lacking systematic analysis or methods for context extension. In this work, we present the first systematic investigation comparing the long-context performance of diffusion LLMs and traditional auto-regressive LLMs. We first identify a unique characteristic of diffusion LLMs, unlike auto-regressive LLMs, they maintain remarkably stable perplexity during direct context extrapolation. Moreover, where auto-regressive models fail outright during the Needle-In-A-Haystack task with context exceeding their pretrained length, we discover diffusion LLMs exhibit a distinct local perception phenomenon, enabling successful retrieval from recent context segments. We explain both phenomena through the lens of Rotary Position Embedding (RoPE) scaling theory. Building on these observations, we propose LongLLaDA, a training-free method that integrates LLaDA with the NTK-based RoPE extrapolation. Our results validate that established extrapolation scaling laws remain effective for extending the context windows of diffusion LLMs. Furthermore, we identify long-context tasks where diffusion LLMs outperform auto-regressive LLMs and others where they fall short. Consequently, this study establishes the first length extrapolation method for diffusion LLMs while providing essential theoretical insights and empirical benchmarks critical for advancing future research on long-context diffusion LLMs. The code is available at https://github.com/OpenMOSS/LongLLaDA.

cs.CL↗