arXiv subjects

Xiao Li

Publications and source records attributed to Xiao Li.

At least 19 recordsLinked to original sources

arXiv2026

Separating Stream Stability from Long-Term Recall in Language Models

Methods for streaming language models are often discussed alongside long-context and memory systems, although they solve different problems. An attention sink can stabilize autoregressive generation over an indefinitely long stream while the model remains unable to use content that has left its recent-token cache. We argue that this distinction should be explicit in system claims and evaluation. We introduce three horizons: the stability horizon, over which predictive behavior remains well behaved; the access horizon, over which past content can still causally affect the output; and the utility horizon, over which a task retains acceptable performance. We show constructively that the stability horizon can be infinite while the access and utility horizons are finite. We then propose ThreeH, an evaluation contract that measures all three horizons under a common state and compute budget. Applying the framework to attention-sink streaming clarifies its strength, constant-memory, stable generation, without treating anchor tokens as semantic memory. The framework exposes roles for cache policies, recurrent state, retrieval, and external memory. Experiments on 128K-token streams, delayed binding recall, and delayed decisions show that attention sinks preserve local modeling but not content beyond the active cache; recurrent and retrieval state extend the semantic horizon.

cs.CL↗

arXiv2026

A Graph Foundation Model for Large-Scale MIMO Detection

Large-scale multiple-input multiple-output (MIMO) detection is fundamental to modern wireless networks but constrained by performance-complexity trade-offs. Existing detectors, whether classical or learning-based, often fall short in either scalability or generalizability across heterogeneous scenarios. To overcome these limitations, we introduce a wireless-native graph foundation model (GFM) tailored for large-scale MIMO detection. The proposed GFM employs a physics-informed hybrid architecture, integrating the local correlation extraction of message passing neural networks with the global attention of graph Transformers, encoding the physical interference patterns from the expectation propagation algorithm. Via extensive pre-training, this synergy enables the learning of a general-purpose detection mapping scalable across antenna dimensions and channel conditions. For rapid downstream deployment, parameter-efficient fine-tuning is leveraged to adapt the GFM to specific non-ideal system regimes with minimal overhead. To enhance inference efficiency, a mixture-of-experts mechanism is embedded at downstream deployment to dynamically activate only the necessary sub-modules. Evaluations show that the proposed GFM consistently outperforms classical detectors and advanced data-driven baselines in accuracy, configuration generality, and cross-scenario transferability across various challenging zero-shot and few-shot conditions.

cs.IT↗

arXiv2026

On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability

We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and additional 51B parameters of n-gram embedding tables held off the accelerator. On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs. Token mixing uses a layer-wise hybrid of Gated DeltaNet (GDN) and global attention, with one full-attention layer in every four; at continued-pretraining time those full-attention layers are replaced by Qwen Sparse Attention (QSA), which scores context at micro-block granularity with a compressed lightweight indexer. The residual stream is widened to four branches and read through an elementwise gate, a design we call the Gated Residual (GR). Capacity is added outside the backbone by a single n-gram embedding layer whose tables are prefetched from host memory. We evaluate every candidate change along three axes: loss together with downstream benchmarks; the cost of the change in training, prefill and decode; and its effect on the optimal hyperparameters and training stability. Loss and downstream accuracy do not always move together: enlarging the n-gram vocabulary lowers loss monotonically while downstream accuracy saturates. The architecture and the Muon optimizer together shift the optimal learning rate and batch size upwards, render batch-size warmup unnecessary, and substantially improve stability under stress tests. Loss, benchmarks, efficiency and stability form one design problem. Solved jointly, they yield a recipe that is simultaneously more efficient, more capable and more stable.

cs.CL↗

arXiv2026

Moir\'e-induced lattice reconstruction at buried atomic interfaces

Atomic reconstruction at twisted two-dimensional interfaces governs many of their emergent optical, electronic, and mechanical properties, including sliding ferroelectricity. Despite recent progress in understanding lattice reconstruction in suspended twisted bilayers, structural changes at van der Waals heterointerfaces between multilayer crystals remain largely unexplored. Here we use multi-slice electron ptychography to non-invasively recover the three-dimensional atomic structure at marginally twisted rhombohedral interfaces between thick transition-metal dichalcogenide crystals. With a position precision of ~3 pm and a depth resolution ~1 nm, we resolve the twist-induced lattice reconstruction field per layer, and the resulting dislocation network at the buried interface. Despite the bulk nature, we observe markedly strong in-plane interfacial reconstruction due to suppression of the out-of-plane bending by outer layers, exceeding predictions from our three-dimensional modelling. Furthermore, we extract the strain tensor evolution during the decay of the reconstruction into the bulk, providing a structural foundation for understanding multi-layer moir\'e systems.

cond-mat.mes-hall↗

arXiv2026

NeuralParker: A Reinforcement Learning Planner for Irregular Parking Environments

Automated parking commonly assumes marked slots and short approach maneuvers. Delivery and service vehicles, however, may need to reach an operator-specified pose in an irregular bounded environment from a distant start. Existing learning-based parking planners often rely on local observations, which can restrict long-range route reasoning. To address this problem, we present NeuralParker, a reinforcement learning-based hybrid planner for arbitrary-pose parking. NeuralParker encodes full-environment obstacle and boundary geometry in a target-relative vertex representation, allowing the policy to retain route-defining context throughout the approach. It further couples a learned curvature--length arc policy with an in-loop terminal ensemble that selects from diverse cubic Hermite connections using a curvature-regularized cost. We also establish factorial and long-range route-choice benchmarks to evaluate planning success and trajectory quality. Experiments on these benchmarks show that NeuralParker achieves higher planning success and better overall trajectory quality than the evaluated baselines, while ablation studies support the benefits of the target-relative global representation and terminal ensemble. Finally, a real-vehicle evaluation confirms that the planner transfers effectively to real delivery-vehicle perception at a working parking site, planning successfully at low computational cost.

cs.RO↗

arXiv2026

Reflex: Enabling Fast and Predictive Vision-Language-Action Models for Reaction-Critical Manipulation

Vision-Language-Action (VLA) models have recently achieved promising performance in robotic manipulation. However, existing benchmarks mainly evaluate generalization on static manipulation tasks and largely overlook dynamic interaction scenarios. To address this gap, we present ReflexBench, a benchmark for reaction-critical manipulation. ReflexBench contains six dynamic tasks and introduces an evaluation framework that decouples simulator stepping from robot control while supporting configurable latency under synchronous and asynchronous inference. Building upon ReflexBench, we propose ReflexVLA, an efficient VLA model designed for reaction-critical manipulation without large-scale robot-data pretraining. ReflexVLA enhances temporal reasoning through latent future prediction and multi-frame temporal fusion within the vision backbone, while reducing deployment latency through batched visual encoding and CUDA Graph replay. Experiments show that ReflexVLA consistently improves dynamic manipulation performance while maintaining competitive accuracy on standard static manipulation benchmarks, and real-world experiments further demonstrate its effectiveness under practical deployment conditions. Project website: https://reflexvla.github.io

cs.RO↗

arXiv2026

Repurposing RGB-based Foundation Model for Depth Estimation on Thermal Images Using Hierarchical Supervision

Depth estimation from thermal images is highly valuable for robotic applications in adverse conditions, such as nighttime and rainy weather. Recent studies have sought to transfer knowledge from RGB-based foundation models to thermal modalities, yet the rich hierarchical representations these models encode remain underutilized. To address this limitation, we propose RGB-HS, a novel framework for thermal-image depth estimation that leverages hierarchical supervision from an RGB-based foundation model. Specifically, we first replace the baseline thermal encoder with a foundational model and introduce a parallel RGB branch that also employs a foundational model as an encoder of the same architecture, taking RGB images as input. The alignment is then performed across multiple levels between the tokens of the two encoders, allowing the thermal student branch to capture both structural precision and semantic abstraction from the RGB teacher branch. Furthermore, we introduce verification to refine the alignment process by weighting tokens from the RGB branch based on RGB image quality. Extensive experiments on the popular benchmark demonstrate that RGB-HS achieves competitive performance and more effectively exploits the representational capacity of RGB-based foundation models for depth estimation on thermal images.

cs.CV↗

arXiv2026

OpenRIS: Democratizing reconfigurable intelligent surfaces for real-world wireless enhancements

Wireless enhancement is critical for next-generation mobile communication systems to realize seamless connectivity, yet traditional network expansion strategies are becoming economically unsustainable. Reconfigurable intelligent surfaces (RISs) provide a promising alternative by improving signal utilization. However, high hardware and deployment costs of advanced RISs limit their large-scale application. Here, we democratize this technology with OpenRIS, an open-source and low-cost platform composed of Lego-like meta-bricks. With digital-twin assistance, these meta-bricks can be flexibly assembled into arbitrary shapes to achieve customized, mass-deployable wireless enhancement without extra power. Experiments and full-wave simulations verify that the discretized OpenRIS achieves consistent performance with the continuous RIS. We further develop a dual-user wireless transmission system and a three-dimensional coverage measurement system to showcase the versatile applicability of OpenRIS in wireless enhancements. As a plug-and-play solution, OpenRIS accelerates the translation of RIS theory into practice and is poised to integrate into infrastructure, reshaping the future wireless world as steel and concrete shape modern cities.

eess.SP↗

arXiv2026

PLS-Calib: A Partial Least Squares Framework for Event Camera and Odometry Calibration under Ground Motion Constraints

Accurate extrinsic rotation calibration between sensors is fundamental to the performance of robotic perception systems. However, most existing calibration techniques rely on full 6-DoF motion to excite all degrees of freedom, which is often infeasible for ground-constrained robots with limited motion capabilities. Recent approaches designed for such restricted settings, such as Canonical Correlation Analysis (CCA)-based methods, suffer from ill-conditioned covariance matrices that lead to numerical instability and suboptimal calibration accuracy. To overcome these limitations, we present a novel rotation calibration framework named PLS-Calib that, for the first time, leverages Partial Least Squares (PLS) regression to model the latent kinematic correlations between asynchronous, heterogeneous sensor streams. Specifically, we apply our method to the calibration of an event camera and an odometry onboard a ground robot. To improve event-based pattern detection, we introduce a polarity-aware event representation, which enhances spatiotemporal contrast in circular calibration targets. Our PLS-based formulation yields a closed-form, stable solution that avoids matrix singularities inherent in CCA-based approaches. Extensive experiments on both synthetic and real-world datasets validate the effectiveness of our approach, demonstrating significant improvements in calibration robustness and accuracy over state-of-the-art methods. This work offers a practical and theoretically grounded solution for rotation calibration in constrained robotic systems and opens up new directions for applying statistical learning techniques in neuromorphic vision.

cs.RO↗

arXiv2026

Interaction-induced sign reversal of the orbital magnetic susceptibility in Chern insulators

It has been well established that the orbital magnetization of interacting electrons can be simply evaluated by applying the single-particle formula to self-consistent HF bands. However, we show that such procedure fails qualitatively for orbital magnetic susceptibility, especially in topological systems: in a Chern-insulating phase of twisted MoTe$_2$, the interaction-induced correction reverses the sign of the susceptibility. This result follows from an algebraic framework that solves the HF problem at finite magnetic field, where noncommuting canonical momenta obstruct a direct calculation: a \emph{reverse Peierls substitution} maps every magnetic-translation-invariant operator to a unique bivariate function, converting the finite-field self-consistency into a function equation that can be expanded systematically in $B$. At first order, this yields a linear equation for the field-induced change $\delta X$ of the Fock potential, leading to an intrinsically interaction-induced susceptibility in addition to a single-particle-like one. The framework reproduces the St\v{r}eda formula for insulators, and an auxiliary-Hilbert-space construction carries it to periodic and moir\'e systems. Finite-field HF calculations in a gapped Dirac model and in the twisted-MoTe$_2$ Chern insulator confirm the theory quantitatively.

cond-mat.str-el↗

arXiv2026

CardioBench: A Real-World Data Benchmark for Evaluating Large Language Models in Clinically Authentic Cardiovascular Care Scenarios

Background: Most medical large language model (LLM) benchmarks focus on examination knowledge or isolated tasks and may not reflect the longitudinal, multimodal, and safety-critical workflow of cardiovascular care. Objective: To develop CardioBench, a real-world benchmark spanning the cardiovascular care continuum, and assess LLM performance across clinical dimensions and specialist tasks. Methods: CardioBench includes 2,263 items from 13 task-specific datasets derived from de-identified cardiovascular records and examination data. Sixteen cardiology physicians conducted annotation and reference construction, followed by cross-review from two senior cardiologists. Seven LLMs generated 15,841 outputs under standardized zero-shot settings. Open-ended tasks were evaluated using key-point coverage and holistic clinical quality, while CardioEthics was scored by accuracy. Results: GPT-5.4 achieved the highest macro-average (62.55) and item-weighted mean (62.19), followed by Gemini 3.1 Pro (59.95) and Qwen 3.6 27B (59.72). GPT-5.4 ranked first in all three dimensions. CardioAuxReport performed best (86.38), whereas CardioECGRead (17.25) and CardioEthics (17.34) were lowest. The largest gaps between holistic clinical quality and key-point coverage occurred in CardioComm (52.71), CardioEmergRescue (52.05), and CardioTreatPlan (48.80). Conclusions: To our knowledge, CardioBench is the largest real-world, multi-task benchmark for LLM evaluation across the cardiovascular care continuum and offers the broadest coverage of clinically authentic cardiology scenarios reported to date. It provides a rigorous framework for identifying model strengths, clinically important omissions, and priorities for future development.

cs.CL↗

arXiv2026

Mage-VL: An Efficient Codec-Native Streaming Multimodal Foundation Model

Standard vision-language models (VLMs) suffer from Moravec's paradox: they excel at complex offline visual reasoning but struggle with simple streaming perception tasks and process them inefficiently. We present Mage-VL, an efficient codec-native streaming foundation model for real-time multimodal understanding and interaction. At its core, our custom tokenizer, Mage-ViT, replaces uniform frame sampling by selectively encoding dynamic, entropy-rich regions using motion vectors and residual energy across sparse anchor (I) and predicted (P) frames. Operating at a 16 x 16 patch level, this reduces visual token consumption by over 75% while preserving spatiotemporal context. Trained from scratch on approximately 560M unlabeled images and 100M unlabeled video frames, Mage-ViT matches or outperforms flagship encoders trained on billions of image-text pairs. We establish AI4AI data pipelines encompassing prompt-code joint optimization for multimodal captioning and AI-driven performance diagnosis to guide training recipes. Furthermore, through a bio-inspired dual-system architecture - a lightweight System 1 event gate and a causal System 2 decoder - Mage-VL enables proactive streaming perception. Extensive evaluations show that Mage-VL-4B matches Qwen3-VL-4B on static tasks while achieving strong gains in video understanding and 2D/3D spatial reasoning, with up to a 3.5x wall-clock inference speedup, and comprehensively surpasses the 15B Phi-4-reasoning-vision baseline. Beyond model artifacts, we deliver seven key empirical findings covering pre-training data efficiency, variable-resolution scaling, codec system acceleration, VideoQA SFT redundancy, motion-spatial synergy, AI4AI data pipelines, and Zero-Vision SFT for multimodal RL.

cs.CV↗

arXiv2026

Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing

Large-scale visual generators are increasingly capable but costly to train, fine-tune, and deploy. We introduce Mage-Flow, a compact 4B-scale generative stack for efficient text-to-image generation and instruction-based image editing. The stack is built from two co-designed components: Mage-VAE, a lightweight high-fidelity latent tokenizer, and a Native-Resolution Multimodal Diffusion Transformer trained with rectified flow matching. Mage-VAE uses one-step diffusion-style encoding and decoding with anchor-latent regularization, preserving the reconstruction quality of strong public VAEs while reducing tokenization cost by more than an order of magnitude. Together with native-resolution packing and stack-level CUDA kernel fusion, the stack supports flexible-resolution training and improves end-to-end training throughput by about $2.5\times$. Built on this foundation, we develop a complete model family with Base, RL-aligned, and Turbo variants for both generation and editing. Diffusion-NFT improves prompt following, text rendering, aesthetic quality, and editing fidelity, while few-step distillation with adversarial perceptual guidance produces 4-step Turbo models for low-latency inference. Despite its compact scale, Mage-Flow and Mage-Flow-Edit achieves competitive performance across standard generation and editing benchmarks. More importantly, the Turbo variants make high-resolution generation and editing practical for interactive use: at $1024^2$ resolution on a single NVIDIA A100 GPU, Mage-Flow-Turbo generates an image in 0.59s, and Mage-Flow-Edit-Turbo edits an image in 1.02s, while maintaining a small memory footprint. These results show that careful tokenizer--backbone--system co-design can deliver strong high-resolution generation and editing within an efficient 4B model family.

cs.CV↗

arXiv2026

Tremerity-Fi: Non-Contact Daily-Life Tremor Severity Assessment by Commercial mmWave Radar

Tremor is a common symptom of neurological diseases. The regular assessment of daily tremors facilitates the evaluation of disease progression and assists clinicians in optimizing treatment strategies. However, current home monitoring solutions have difficulty in dealing with user cooperation, privacy concerns, environmental interference, and system generalization, leading to feasibility concerns in activities of daily living (ADL). To this end, we propose Tremerity-Fi, a non-contact and privacy-friendly tremor severity assessment system based on mmWave radar. To realize Tremerity-Fi, we first design an adaptive beamforming algorithm to accurately identify useful but weak signals from numerous reflections captured in the environment. Second, unlike primary reflections commonly used in mmWave sensing, we leverage multipath reflections that carry useful information about the target's motion, even though they are generally considered harmful, to help reconstruct hand signals and improve sensing performance. Furthermore, we propose an unsupervised domain adaptation algorithm to improve the ability to adapt to unseen environments and users. We collect a diverse dataset of 5 patients and 25 healthy subjects in 3 scenarios, such as offices, homes, and hospitals. Extensive experiments show that our system achieves 94.51% accuracy in tremor detection, about 5 higher than the SOTA mmWave radar method, and 89.13% in tremor severity assessment, demonstrating its sufficient potential as a tremor monitoring assistant for patients with neurological diseases.

eess.SP↗

arXiv2026

Harness VLA: Steering Frozen VLAs into Reliable Manipulation Primitives via Memory-Guided Agents

Language-conditioned manipulation requires both precise contact-rich control and robust reasoning over language, scenes, and long horizons. End-to-end Vision-Language-Action (VLA) models provide strong local visuomotor skills, but they are trained on in-distribution task trajectories and often fail under deployment perturbations such as semantic retargeting, goal re-binding, spatial-layout shifts, and unstable local contacts. LLM coding agents provide complementary semantic and compositional reasoning, but purely analytic primitives struggle with irregular grasping, constrained placement, and articulated-object interaction. We present Harness VLA, a memory-augmented agentic framework that exposes a frozen VLA as a retryable contact-rich primitive and composes it with a small fixed library of analytic primitives for grounding, staging, transport, navigation, and release. Rather than expanding the skill library, the harness learns the operating range of these fixed primitives from task-specific execution traces, global success rules, and failure models. By lifting semantic re-grounding, non-contact execution, and VLA re-staging to the planner while reserving the frozen VLA for local contact-rich phases, Harness VLA extends pretrained VLAs beyond their original trajectory distribution without finetuning. Across perturbed tabletop, household kitchen, and clean-to-randomized bimanual manipulation, Harness VLA improves over the strongest relevant baselines by 38.6 and 25.4 percentage points on LIBERO-Pro and RoboCasa365, respectively, and reaches 58.4% on RoboTwin C2R. Code is available at https://github.com/RLinf/RPent.

cs.RO↗

arXiv2026

Signatures of Two Distinct Epochs of FRB 20240114A from January to August 2024 Based on its Energy and Waiting Time Analysis

A comprehensive analysis of the energy and waiting time distributions of the bursts from FRB 20240114A detected by the Five-hundred-meter Aperture Spherical Radio Telescope between 28 January and 29 August 2024 is presented. For the full sample, its energy distribution cannot be fitted with the simple power-law (SPL),bent power-law (BPL), thresholded power-law (TPL) or Band function models, and its waiting time distribution excluding intervals shorter than 0.5 s cannot be fitted with the Poisson or Weibull models. Nevertheless, for the subsamples with more than 50 bursts in single-day observations, their energy distributions can be fitted with the BPL or TPL models, and their waiting time distributions are better described by a Weibull model. It is noted that the best-fitting BPL parameter $\beta$ is approximately invariant within the epochs before and after 21 March 2024, with an average of $\bar \beta_b = 1.006 \pm 0.074$ and $\bar \beta_a = 1.236 \pm 0.183$ (one standard deviation), respectively. Most subsamples from the later epoch have a smaller burst rate parameter $r$ in the Weibull model than those from the earlier epoch. The majority of bursts with $E>10^{39}$ erg occurred in the earlier epoch. The energy distributions in the high-energy range ($> 6\times10^{37}$ erg) differ significantly between the two epochs, and power-law fits to $dN/dE$ yield indices of $-1.97_{-0.02}^{+0.02}$ and $-2.34_{-0.06}^{+0.06}$, respectively. The median of the waiting time distribution of the later epoch is larger than that in the earlier epoch. These results suggest that the two epochs may be dominated by different types of bursts, possibly attributed to changes in the physical properties of the emission region.

astro-ph.HE↗

arXiv2026

Erase-then-Delta Attention: Decoupling Erase and Write Addresses in Delta-Rule Linear Attention

Delta-rule linear attention improves recurrent memory updates by correcting what is already stored at the current write address before writing new content. However, the active correction is still anchored to that same write address. As a result, stale information stored at a different address cannot be actively removed before new content is written elsewhere. We propose Erase-then-Delta Attention (EDA), a memory update rule that decouples where to erase from where to write. The key insight is that recurrent memory models should not only correct the current write, but also selectively suppress outdated memory at an independently chosen address. Concretely, our method first applies a targeted erase step along a learned erase direction, and then performs the standard delta-style corrective write along the current write direction. This preserves the corrective behavior of delta-rule updates while expanding their memory-management capacity. Language-model pretraining experiments across dense 2.5B and MoE 25B-A2.8B model families show that EDA performs best in both settings. The gain persists after 80B-token long-context midtraining of the MoE models, where EDA also performs best in long-context evaluations from 4k to 128k contexts. A compact update analysis and memory-state probes suggest why: EDA keeps the delta-rule corrective write intact while allocating an additional cleanup path most strongly when passive decay is weak. These results suggest that recurrent memory models should decide not only what to write, but also what stale information to erase and where.

cs.CL↗

arXiv2026

Load Testing for Machine Learning Model Serving Systems at Scale

Machine learning (ML) model serving has become a dominant consumer of GPU infrastructure, yet capacity planning in these systems remains largely ad hoc. Under-provisioning leads to service-level objective (SLO) violations and production incidents, while over-provisioning results in substantial resource waste. This paper presents \sys, an industrial load testing framework for ML serving systems that systematically estimates serving capacity through an adaptive, feedback-driven search strategy. The approach leverages real-time performance signals, incorporating dampening, spike tolerance, and convergence detection to efficiently identify maximum sustainable throughput under SLO constraints. We evaluate \sys through a longitudinal analysis of 14 industrial case studies spanning four ML architecture classes: recommendation, ranking, vision, and NLP. This study demonstrates that systematic load testing leads to substantial improvements in GPU resource efficiency and operational reliability. Prior to adopting \sys, a significant fraction of model launches were under-provisioned, resulting in recurring incidents; these issues were substantially reduced after deployment. Our results show that ML-specific design decisions are critical to accurate capacity estimation: workload calibration using recorded traffic reduces estimation error from approximately 30\% to 2--6\%, while proper warmup handling yields a 22.2\% improvement in accuracy. Further analysis reveals key factors influencing prediction error, including model size and co-location effects. This paper distills six lessons and derive architectural guidelines for ML load testing, offering actionable insights for building reliable and efficient ML serving systems.

cs.LG↗