arXiv subjects

Rui Zhou

Publications and source records attributed to Rui Zhou.

At least 37 records · Page 2Linked to original sources

arXiv2026

Fora: From Weight-Space to Function-Space Protection in Capability-Preserving Fine-Tuning

Full fine-tuning adapts large language models to new tasks but can erode capabilities they already possess. Existing remedies protect through proxies such as parameter distances, importance penalties, output matching, or dominant singular directions of the weights, but none directly asks which activation directions the preserved capability relies on. We argue that a capability is characterized more faithfully by the activation subspace it induces than by the singular geometry of the weight matrix, and develop function-space protection, instantiated as FORA (Function-space Orthogonal Residual Adaptation). From label-free calibration inputs, FORA estimates, per layer, the principal directions $Q$ of the input-activation covariance and forms a right projector $P_Q = I - QQ^T$. Paired with a left projector $P_U$ from the weight SVD, the update is $ΔW = P_U M P_Q + U_2 D_δ V_2^T$: a high-capacity branch structurally barred from reading capability-relevant function directions, plus a narrow spectral channel for controlled plasticity. The construction extends to parameter-efficient adaptation via $M \to (α/r) BA$. Across three settings on Qwen3-1.7B, including COGS and GSM8K learned while preserving translation and translation learned while preserving math, FORA consistently improves preservation over weight-space projection and standard regularization, with only a small new-task trade-off in the math-preservation setting. A controlled ablation isolating the projection source shows that the advantage comes not from projection itself, but from projecting onto capability-derived rather than weight-derived directions. Code is available at https://github.com/zrui239/FORA.

cs.LG↗

arXiv2026

Metamagnetism in UTe2: the roles of itinerancy and localization

The metamagnetic transition in UTe$_2$ plays a key role in stabilizing two enigmatic field-induced superconducting phases. One of these phases (SC2) is truncated by the transition, lying directly below it, while the other (SC3) sits predominantly above it and appears to be stabilized because of it. While numerous pulsed field studies have examined this transition, comparatively few steady field experiments have investigated it. Here we report a suite of measurements of metamgnetism in UTe$_2$, at ambient pressure by torque magnetometry and extraction magnetometry techniques, and of the magnetoconductance under pressure. Our steady field measurements resolve a complex sub-structure within the transition, with separate features that possess different temperature evolutions, pointing to distinct contributions from itinerant and localized moments. The itinerant contribution might relate to a possible spin-density wave state. We theoretically model the evolution of Kondo and RKKY interactions and propose that the SC2 state is stabilized under pressure due to the collapse of magnetic anisotropy, leading to an enhancement of longitudinal spin fluctuations along the hard $b$ axis, which are pair-forming in the $p$-wave channel.

cond-mat.str-el↗

arXiv2026

GreenRFM: Learning a resource-efficient radiology vision-language foundation model via supervision-centric pre-training

Radiology foundation models (RFMs) have largely inherited the scale-first recipe of natural-image vision--language pre-training. This recipe is difficult to deploy in 3D radiology, where training corpora are smaller, reports vary across institutions, and receiving hospitals often need local adaptation under privacy and compute constraints. We ask whether routine radiology reports can instead be converted into auditable diagnostic supervision that shapes the image encoder, text encoder, aligned space, and local-adaptation procedure. We develop GreenRFM, a supervision-centric pre-training framework organized around four empirical principles: More distilled, Ubiquitous, Semantic-enforcing, and Task-aligning (MUST) supervision. These principles convert noisy reports into structured diagnostic signals and use them to learn discriminative unimodal encoders plus an aligned image--text space for diagnosis-centered multimodal use. GreenRFM requires 24 GPU-hours on a single 24GB GPU (lightweight variant: 6GB VRAM, 4~hours) and reaches a zero-shot CT-RATE AUC of 84.8. Evaluations using more than 200,000 volumes from six institutions and two modalities show transfer to private clinical cohorts and to musculoskeletal MRI. On a local institutional cohort, computationally feasible retraining raises macro-AUC from 70.5 to 82.1. The aligned space also improves hepatocellular-carcinoma microvascular-invasion prediction and trans-arterial chemoembolization response analysis over established clinical scores. These results support supervision-centric pre-training as a practical route to resource-efficient, locally adaptable, diagnosis-centered radiology vision--language representations.

cs.CV↗

arXiv2026

ESPO: Early-Stopping Proximal Policy Optimization

When a large language model under reinforcement learning commits a wrong reasoning step early in a trajectory, standard algorithms force it to keep generating until the maximum horizon, spending compute on tokens that never receive positive reward and polluting advantage estimates with post-failure noise. We propose ESPO (Early-Stopping Proximal Policy Optimization), which detects trajectory failure on-the-fly and terminates rollouts early. At each generation step, ESPO computes a surrogate regret using only the logits already computed during sampling, and terminates when the smoothed cumulative regret significantly exceeds its estimated values. Truncated trajectories are treated as absorbing failure states with a terminal reward, concentrating negative temporal-difference (TD) errors near the detected failure step without any additional reward model or human annotation. On DeepSeek-R1-Distill-Qwen-7B trained for mathematical reasoning, ESPO surpasses PPO on AIME~2024 (46.28% vs. 45.25%), AMC~2023 (85.83% vs. 82.94%), and MATH-500 (87.42% vs. 85.43%), while saving more than 20% rollout tokens cumulatively.

cs.LG↗

arXiv2026

TwinRL: Digital Twin-Driven Reinforcement Learning for Real-World Robotic Manipulation

Despite strong generalization capabilities, Vision-Language-Action (VLA) models remain constrained by the high cost of expert demonstrations and limited real-world interaction. While online reinforcement learning (RL) has shown promise, its application to real-world VLA manipulation is hindered by low exploration efficiency and restricted exploration coverage. Through systematic real-world experiments, we observe that the effective exploration space of online RL is largely constrained by the trajectory distribution induced during supervised fine-tuning (SFT). Motivated by this observation, we propose TwinRL, a digital twin-real-world collaborative post-training framework that expands and guides RL exploration for VLA models through three stages: SFT warm-up, twin RL warm-up, and real-world RL. TwinRL first reconstructs a high-fidelity digital twin from smartphone-captured scenes. During the SFT stage, we introduce an exploration space expansion strategy that expands the support of the trajectory distribution beyond real demonstrations, reshaping the exploration space for more effective RL. Rather than treating the twin as a data augmentation tool, we propose a twin RL warm-up strategy that enables it to act as an exploration guide for real-world RL. Specifically, TwinRL performs efficient parallel RL in the digital twin to generate interactive trajectories that populate the replay buffer and stabilize subsequent real-world RL learning. This process also identifies failure-prone yet informative configurations, enabling targeted human-in-the-loop rollouts to further improve on-robot efficiency. Across four tasks, TwinRL achieves near-100% success in both in-distribution and out-of-distribution regions, delivering over 30% faster convergence than prior real-world RL methods with only 20 minutes of on-robot interaction.

cs.RO↗

arXiv2026

How to augment cosmic shear measurements with radio polarimetry of galaxies?

The integral polarization of spiral galaxies in the radio band has been proposed as a new tracer of the intrinsic galaxy shape that augments lensing shear measurements. We revisit the method of shear estimation in this context. We introduce a new statistical model in which galaxy shape and polarization are Gaussian random variables with their covariance characterizing the quality of polarization-shape alignment. Applying the principle of likelihood maximization, we then analytically derive unbiased, minimal-variance estimators, which allow to simultaneously estimate gravitational shear, intrinsic shape alignment and line-of-sight polarization rotation, all at once and accurate to first order in these three effects. New to the literature, our estimators have the merits of being free of biases, robust in situations of few galaxies or poor polarization-shape alignment, allowing analytic reconstruction noise covariance, and minimizing uncertainties in power spectrum estimation, thus resolving conceptual issues of the existing estimation methods. This new analytic framework is generally applicable to future research that exploits the polarization-shape alignment effect of galaxies.

astro-ph.CO↗

arXiv2026

Discover Fast Power Allocation Solution for Multi-Target Tracking via AlphaEvolve Evolution

Efficient radar resource allocation is a fundamental yet computationally challenging problem, as optimal solutions typically require iterative optimization with high complexity. Motivated by the need for real-time scheduling, robust generalization, and low data dependency, this paper proposes a novel paradigm that leverages large language model (LLM)-guided evolutionary search (AlphaEvolve) to autonomously discover a closed-form power allocation solution for multi-target tracking. The approach encodes high-dimensional radar states into physically inspired features, then evolves a compact and interpretable scoring function, which is transformed to feasible power allocations via a deterministic constraint-satisfying transformation. Extensive experiments demonstrate that the discovered closed-form solution achieves near-optimal tracking accuracy (average relative performance loss of only $1.51\%$), reliable generalization across diverse scenarios and target counts, and over three orders of magnitude speedup compared to conventional iterative solvers. These results highlight the potential of LLM-guided symbolic search to revolutionize not only radar resource management but also broader classes of engineering optimization problems.

eess.SP↗

arXiv2026

Electrically-controllable superconducting memory effect in UTe2

If a computer could be assembled from superconducting components, the energy efficiency would far surpass that of conventional electronics. Historic research efforts towards this goal yielded pivotal breakthroughs in the development and discovery of scanning tunnelling microscopy and high temperature superconductivity. Although recent strides have been taken in advancing superconducting diode and switching technologies, harnessing read/writeable memory functionality in superconducting platforms has remained challenging. Here we show that bulk single crystal specimens of the triplet superconductor candidate uranium ditelluride (UTe$_2$) possess such properties. Upon applying a magnetic field to access an intermediate regime straddling two distinct superconducting phases, we find that direct current pulses can push the material in and out of a metastable state possessing an enhanced critical current $J_c$. This switching is controllable by the strength and duration of the stimuli, with the system `remembering' whether it is in the high or low $J_c$ state for extended periods. We interpret this to be due to competition between two distinct vortex species, which can be perturbatively pushed into a non-equilibrium high-disorder configuration with stronger pinning forces and thus higher $J_c$. Rather than requiring proximate magnetic or semiconducting interfaces, this memory functionality appears to be an intrinsic property of UTe$_2$ rooted in the superconducting order itself. Our findings underscore the rich complexity of quantum vortex matter, and demonstrate the viability of engineering a new class of superconducting memory elements with ultralow-power switching.

cond-mat.supr-con↗

arXiv2026

Bulk superconductivity up to 96 K in pressurized nickelate single crystals

Recently, the Ruddlesden-Popper bilayer nickelate $La_3Ni_2O_7$ has emerged as a superconductor with a transition temperature ($T_c$) of approximately 80 K above 14 GPa (Refs. 1-3). Achieving higher $T_c$ in nickelate superconductors, along with the synthesis of reproducible high-quality single crystals without relying on high-oxygen-pressure growth conditions, remains a significant challenge$^{[4-7]}$. Here we report superconductivity up to 96 K under high pressure in bilayer nickelate single crystals synthesized at ambient pressure. Energy-dispersive spectroscopy, single-crystal X-ray diffraction, nuclear quadrupole resonance and scanning transmission electron microscopy evidenced high crystal quality of the flux-grown $La_2SmNi_2O_{7-δ}$ single crystals. $La_2SmNi_2O_7$ exhibits clear bulk superconductivity, including zero resistivity ($T_{c,max}^{onset}$ = 92 K and $T_{c,max}^{zero}$ = 73 K at 21.6 GPa) and the Meissner effect ($T_c$= 60 K at 20.6 GPa). A low-temperature high-pressure structural study indicates that both monoclinic and tetragonal structures can support superconductivity in this bilayer nickelate. Furthermore, we established a correlation between higher $T_c$ under high pressures and larger in-plane lattice distortion under ambient conditions, corroborated by observing even higher $T_c^{onset}$ of 96 K in $La_{1.57}Sm_{1.43}Ni_2O_{7-δ}$. This study overcomes key limitations in growing nickelate superconductor crystals, resolves the crystal structure in the superconducting state and demonstrates an effective pathway towards achieving higher $T_c$.

cond-mat.supr-con↗

arXiv2026

Memory Over Maps: 3D Object Localization Without Reconstruction

Target localization is a prerequisite for embodied tasks such as navigation and manipulation. Conventional approaches rely on constructing explicit 3D scene representations to enable target localization, such as point clouds, voxel grids, or scene graphs. While effective, these pipelines incur substantial mapping time, storage overhead, and scalability limitations. Recent advances in vision-language models suggest that rich semantic reasoning can be performed directly on 2D observations, raising a fundamental question: is a complete 3D scene reconstruction necessary for object localization? In this work, we revisit object localization and propose a map-free pipeline that stores only posed RGB-D keyframes as a lightweight visual memory--without constructing any global 3D representation of the scene. At query time, our method retrieves candidate views, re-ranks them with a vision-language model, and constructs a sparse, on-demand 3D estimate of the queried target through depth backprojection and multi-view fusion. Compared to reconstruction-based pipelines, this design drastically reduces preprocessing cost, enabling scene indexing that is over two orders of magnitude faster to build while using substantially less storage. We further validate the localized targets on downstream object-goal navigation tasks. Despite requiring no task-specific training, our approach achieves strong performance across multiple benchmarks, demonstrating that direct reasoning over image-based scene memory can effectively replace dense 3D reconstruction for object-centric robot navigation. Project page: https://ruizhou-cn.github.io/memory-over-maps/

cs.RO↗

arXiv2026

Driving risk emerges from the required two-dimensional joint evasive acceleration

Most autonomous driving safety benchmarks use time-to-collision (TTC) to assess risk and guide safe behaviour. However, TTC-based methods treat risk as a one-dimensional closing problem, despite the inherently two-dimensional nature of collision avoidance, and therefore cannot faithfully capture risk or its evolution over time. Here, we report evasive acceleration (EA), a hyperparameter-free and physically interpretable two-dimensional paradigm for risk quantification. By evaluating all possible directions of collision avoidance, EA defines risk as the minimum magnitude of a constant relative acceleration vector required to alter the relative motion and make the interaction collision-free. Using interaction data from five open datasets and more than 600 real crashes, we derive percentile-based warning thresholds and show that EA provides the earliest statistically significant warning across all thresholds. Moreover, EA provides the best discrimination of eventual collision outcomes and improves information retention by 54.2-241.4% over all compared baselines. Adding EA to existing methods yields 17.5-95.5 times more information gain than adding existing methods to EA, indicating that EA captures much of the outcome-relevant information in existing methods while contributing substantial additional nonredundant information. Overall, EA better captures the structure of collision risk and provides a foundation for next-generation autonomous driving systems.

cs.RO↗

arXiv2026

GenHSI: Controllable Generation of Human-Scene Interaction Videos

Large-scale pre-trained video diffusion models have exhibited remarkable capabilities in diverse video generation. However, existing solutions face several challenges in generating long videos with rich human-scene interactions (HSI), including unrealistic dynamics and affordance, lack of subject identity preservation, and the need for expensive training. To this end, we propose GenHSI, a training-free method for controllable generation of long HSI videos with 3D awareness. Taking inspiration from movie animation, we subdivide the video synthesis into three stages: (1) script writing, (2) pre-visualization, and (3) animation. Given an image of a scene and a character with a user description, we use these three stages to generate long videos that preserve human identity and provide rich and plausible HSI. Script writing converts a complex text prompt involving a chain of HSI into simple atomic actions that are used in the pre-visualization stage to generate 3D keyframes. To synthesize plausible human interaction poses in 3D keyframes, we utilize pre-trained 2D inpainting diffusion models to generate plausible 2D human interactions based on view canonicalization, which eliminates the need for multi-view fitting in previous works. We then extend these interactions to 3D using robust iterative optimization, informed by contact cues and reasoning from VLMs. Prompted by these 3D keyframes, the pretrained video diffusion models can better generate consistent long videos with plausible dynamics and affordance in a 3D-aware manner. We are the first to synthesize a long video sequence with a chain of HSI actions without training based on the image references of the scene and character. Experiments demonstrate that our method can generate HSI videos that effectively preserve scene content and character identity with plausible human-scene interaction from a single image scene.

cs.CV↗

arXiv2026

Seeing but Not Thinking: Routing Distraction in Multimodal Mixture-of-Experts

Multimodal Mixture-of-Experts (MoE) models have achieved remarkable performance on vision-language tasks. However, we identify a puzzling phenomenon termed Seeing but Not Thinking: models accurately perceive image content yet fail in subsequent reasoning, while correctly solving identical problems presented as pure text. Through systematic analysis, we first verify that cross-modal semantic sharing exists in MoE architectures, ruling out semantic alignment failure as the sole explanation. We then reveal that visual experts and domain experts exhibit layer-wise separation, with image inputs inducing significant routing divergence from text inputs in middle layers where domain experts concentrate. Based on these findings, we propose the Routing Distraction hypothesis: when processing visual inputs, the routing mechanism fails to adequately activate task-relevant reasoning experts. To validate this hypothesis, we design a routing-guided intervention method that enhances domain expert activation. Experiments on three multimodal MoE models across six benchmarks demonstrate consistent improvements, with gains of up to 3.17% on complex visual reasoning tasks. Our analysis further reveals that domain expert identification locates cognitive functions rather than sample-specific solutions, enabling effective transfer across tasks with different information structures.

cs.CV↗

arXiv2026

Cascade of Spin Liquids in a Bilayer Triangular-lattice Antiferromagnet Rb_2Co_2(SeO_3)_3

In frustrated Ising magnets, classical spin liquids (CSLs) with macroscopic ground-state degeneracy can survive against conventional magnetic order, as exemplified by systems on triangular, kagome and pyrochlore lattices at zero field. Here we report the discovery of a high-field route toward spin liquids in a bilayer triangular lattice antiferromagnet, Rb$_2$Co$_2$(SeO$_3$)$_3$. We demonstrate that a cascade of CSLs -- characterized by doubly degenerate one-up-one-down local spin configurations and a residual entropy of 1/2(1-M/M_s)Rln2 per mole -- emerges through field-controlled dilution of Ising dimers. Owing to the interplay of intra- and inter-layer interactions, these CSLs are further stabilized by lattice symmetry breaking at fractional magnetization plateaus. Such field-induced spin liquids can be understood as a consequence of generalized ice rules, analogous to those governing in pyrochlore antiferromagnets. In particular, the 5/6-plateau state is a candidate quantum spin liquid. Our results thereby establish a new pathway for exploring diverse spin liquid states across both classical and quantum regimes.

cond-mat.str-el↗

arXiv2026

DIET: Learning to Distill Dataset Continually for Recommender Systems

Modern deep recommender models are trained under a continual learning paradigm, relying on massive and continuously growing streaming behavioral logs. In large-scale platforms, retraining models on full historical data for architecture comparison or iteration is prohibitively expensive, severely slowing down model development. This challenge calls for data-efficient approaches that can faithfully approximate full-data training behavior without repeatedly processing the entire evolving data stream. We formulate this problem as \emph{streaming dataset distillation for recommender systems} and propose \textbf{DIET}, a unified framework that maintains a compact distilled dataset which evolves alongside streaming data while preserving training-critical signals. Unlike existing dataset distillation methods that construct a static distilled set, DIET models distilled data as an evolving training memory and updates it in a stage-wise manner to remain aligned with long-term training dynamics. DIET enables effective continual distillation through principled initialization from influential samples and selective updates guided by influence-aware memory addressing within a bi-level optimization framework. Experiments on large-scale recommendation benchmarks demonstrate that DIET compresses training data to as little as \textbf{1-2\%} of the original size while preserving performance trends consistent with full-data training, reducing model iteration cost by up to \textbf{60$\times$}. Moreover, the distilled datasets produced by DIET generalize well across different model architectures, highlighting streaming dataset distillation as a scalable and reusable data foundation for recommender system development.

cs.IR↗

arXiv2026

Video2Act: A Dual-System Video Diffusion Policy with Robotic Spatio-Motional Modeling

Robust perception and dynamics modeling are fundamental to real-world robotic policy learning. Recent methods employ video diffusion models (VDMs) to enhance robotic policies, improving their understanding and modeling of the physical world. However, existing approaches overlook the coherent and physically consistent motion representations inherently encoded across frames in VDMs. To this end, we propose Video2Act, a framework that efficiently guides robotic action learning by explicitly integrating spatial and motion-aware representations. Building on the inherent representations of VDMs, we extract foreground boundaries and inter-frame motion variations while filtering out background noise and task-irrelevant biases. These refined representations are then used as additional conditioning inputs to a diffusion transformer (DiT) action head, enabling it to reason about what to manipulate and how to move. To mitigate inference inefficiency, we propose an asynchronous dual-system design, where the VDM functions as the slow System 2 and the DiT head as the fast System 1, working collaboratively to generate adaptive actions. By providing motion-aware conditions to System 1, Video2Act maintains stable manipulation even with low-frequency updates from the VDM. For evaluation, Video2Act surpasses previous state-of-the-art VLA methods by 7.7% in simulation and 21.7% in real-world tasks in terms of average success rate, further exhibiting strong generalization capabilities.

cs.RO↗

arXiv2026

GOMA: Geometrically Optimal Mapping via Analytical Modeling for Spatial Accelerators

General matrix multiplication (GEMM) on spatial accelerators is highly sensitive to mapping choices in both execution efficiency and energy consumption. However, the mapping space exhibits combinatorial explosion, which makes it extremely challenging to obtain optimal mappings within an acceptable time budget. Existing approaches typically face challenges: They often lack global-optimality guarantees and become prohibitively slow as the mapping space grows. To address these limitations, we propose \textsc{GOMA}, a geometric-abstraction-based, globally optimal GEMM mapping framework via analytical modeling, which achieves efficient solving while guaranteeing optimality. \textsc{GOMA} introduces, from first principles, a geometric abstraction for GEMM mapping, yielding an exact analytical energy objective with $O(1)$ evaluation for any given mapping. The objective is highly accurate. \textsc{GOMA} then formulates mapping selection as an integer optimization problem under hardware and mapping constraints, using the analytical energy model as the objective to automate mapping search. \textsc{GOMA} can quickly compute a global-optimal mapping for any (GEMM workload, target hardware) pair, achieving this for the first time in mapping space exploration. Experiments confirm that across representative accelerators and large language model prefill workloads, \textsc{GOMA} improves the energy--delay product (EDP) by $2.24$--$4.24\times$ over SOTA mappers, while accelerating time-to-solution by $3.83$--$73.6\times$.

cs.AR↗

arXiv2026

Observation of Kondo hybridization wave in UTe2

Condensed matter systems with strong electronic correlations often manifest a variety of intertwined ordered phases of charge, spin, orbital and other degrees of freedom. As a prototypical strongly correlated electronic system, the Kondo lattice provides fertile soil for many fascinating quantum states, including quantum criticality, unconventional superconductivity, hidden order and topological Kondo insulator/semimetal. The foundation of Kondo physics lies in the hybridization between localized moments and itinerant electrons. Generally, the evolution of Kondo hybridization is characterized as a broad crossover rather than a phase transition. Thus far, an ordered hybridization phase has not been observed. Here, we use scanning tunneling microscopy (STM) to identify a translational-symmetry-breaking order of Kondo hybridization wave(KHW) for the first time on the surface of the spin-triplet heavy-fermion superconductor UTe2. The unprecedented phase of KHW manifests as a periodically modulated Fano lattice, accompanied by a commensurate charge density wave (CDW) and a pronounced energy gap opening near the Fermi level. This KHW-imprinted CDW has an intriguing real-space texture of complementary occupation of the heavy f and conduction charges, thereby forming a Kondo superlattice. The KHW is coexistent with superconductivity in UTe2, which may provide valuable insight into its controversial spin-triplet pairing symmetry and the underlying mechanism. Our first experimental evidence for an ordered hybridization state potentially sheds new light on the strong correlation physics of Kondo lattice system.

cond-mat.str-el↗