arXiv subjects

Yufan Xu

Publications and source records attributed to Yufan Xu.

18 recordsLinked to original sources

arXiv2026

SPEAR: Selection-aware Personalized End-to-end Adaptive Rewriting and Retrieval for Community Search

Query reformulation bridges user intent and retrieval in e-commerce search, yet production systems optimize rewrite quality and retrieval effectiveness separately, leaving the two stages structurally misaligned. Path-based architectures unify them end-to-end but were designed for personalization, where relevance is not an explicit constraint-search additionally requires the rewrite to remain faithful to the user's stated query intent. Transplanted directly, these models learn a shortcut we term the generic-word dominance effect: they favor generic rewrites that score well on paths but drift from query intent. To address this, we propose SPEAR (Selection-aware Personalized End-to-end Adaptive Rewriting and Retrieval), which integrates three components that each target one failure mode: (1) a dual-embedding backbone with auxiliary loss and gradient isolation that shields recall-side semantics from being eroded by CTR-driven ranking signals; (2) a multiplicative gating aggregator that lets a rewrite score high only when both its confidence and item relevance are strong, eliminating the generic-word shortcut; (3) a Dynamic Rewrite Selector that jointly generates request-specific rewrite weights and user-query-conditioned scale and bias terms, allowing both rewrite preference and relevance calibration to adapt to each request. Offline evaluation on 100K held-out industrial search sessions shows that the proposed framework improves rewrite semantic similarity@10 by +18.2 and click recall@10 by +99.5 over the production baseline. In online A/B testing, SPEAR achieves +0.259 in query-view CTR and +0.733 in average reading depth, confirming that improved rewrite selection translates into stronger retrieval and deeper user engagement. The proposed SPEAR system has been fully deployed in Dewu's community search platform since 2025. Our code is available at https://github.com/mallocagi1-cell/spear.

cs.IR↗

arXiv2026

Extension of a multi-region free-surface MHD solver beyond the inductionless approximation

Free-surface liquid metal flows are a leading candidate for the plasma-facing components of future fusion reactors, but existing transient, three-dimensional, free-surface MHD solvers rely on the inductionless approximation, in which the induced magnetic field is neglected. This paper extends the open-source solver FreeMHD [B. Wynne et al., Phys. Plasmas 32, 013907 (2025)] beyond that approximation, resolving the induced field self-consistently with a vector-potential formulation that enforces $\nabla\cdot\boldsymbol{B}=0$ by construction while preserving the original multi-region, two-phase framework. It is verified against the analytical Shercliff and Hunt duct flows, against flows driven by a time-varying applied field, and against the deformation of a free liquid metal jet crossing a non-uniform field, and validated against free-surface height measurements from the LMX-U experiment. The experiment validates the overall free-surface solution rather than finite-$R_m$ effects, which the transient-field cases verify. To our knowledge this is the first open-source, fully three-dimensional free-surface liquid metal solver to resolve the evolution of the induced magnetic field, providing a basis for modeling the finite magnetic Reynolds number conditions expected in large-scale, transient fusion events.

physics.comp-ph↗

arXiv2026

Lithium Experimental Application Platform (LEAP): Secondary-Containment Architecture for Flowing Liquid Lithium in Fusion Systems

Flowing liquid lithium is a promising fusion technology because it can provide a renewable Plasma-Facing Component (PFC) surface, modify recycling, support power exhaust, and potentially connect plasma-facing components with fuel recovery. Its deployment, however, is limited by the need to manage chemical reactivity, fire and aerosol hazards, inert gas operation, maintainability, and rapid experimental iteration. This paper develops a semi-quantitative hazard complexity framework for selecting secondary containment architectures for flowing liquid lithium systems. The framework is applied to six representative containment scenarios and to the Lithium Experimental Application Platform (LEAP) at Princeton Plasma Physics Laboratory. LEAP is under construction with a modular, room-scale argon gloveroom as an inert secondary containment boundary for a staged flowing lithium program with heating, diagnostics, magnetic field exposure, and future device interface capability. The analysis shows that an inert, airtight secondary enclosure without scrubbers around a liquid lithium loop provides a practical balance between hazard reduction and facility complexity, as defined by the design requirements. The resulting architecture offers a deployable path for lithium PFC development and a transferable design logic for other reactive or conductive liquid metal systems.

physics.app-ph↗

arXiv2026

SDiT: Semantic Region-Adaptive for Diffusion Transformers

Diffusion Transformers (DiTs) achieve state-of-the-art performance in text-to-image synthesis but remain computationally expensive due to the iterative nature of denoising and the quadratic cost of global attention. In this work, we observe that denoising dynamics are spatially non-uniform-background regions converge rapidly while edges and textured areas evolve much more actively. Building on this insight, we propose SDiT, a Semantic Region-Adaptive Diffusion Transformer that allocates computation according to regional complexity. SDiT introduces a training-free framework combining (1) semantic-aware clustering via fast Quickshift-based segmentation, (2) complexity-driven regional scheduling to selectively update informative areas, and (3) boundary-aware refinement to maintain spatial coherence. Without any model retraining or architectural modification, SDiT achieves up to 3.0x acceleration while preserving nearly identical perceptual and semantic quality to full-attention inference.

cs.CV↗

arXiv2025

RLHFSpec: Breaking the Efficiency Bottleneck in RLHF Training via Adaptive Drafting

Reinforcement Learning from Human Feedback (RLHF) is an important fine-tuning technique for large language models (LLMs) and comprises three stages: generation, inference, and training. The generation stage generates samples that are then used to infer learnable experiences for training. We observe that the generation stage is the bottleneck of the entire execution process and consider it a key point for optimization. Specifically, we realize the first attempt to integrate speculative decoding into the RLHF generation stage and propose RLHFSpec, an RLHF system that accelerates generation execution with efficient speculative decoding and sample reallocation. To fully exploit the performance potential provided by speculative decoding, especially dealing with the dynamic workload of the generation stage, RLHFSpec proposes a workload-aware drafting strategy selection mechanism, which selects the near-optimal strategy by jointly considering the verification cost and the number of accepted tokens. Moreover, RLHFSpec also proposes sample reallocation to fully utilize the GPU resources, and optimizes it with an efficient sample migration mechanism. The experimental results show that the RLHFSpec can achieve higher throughput in the generation stage compared to state-of-the-art works. Moreover, due to the effective alleviation of the generation bottleneck, RLHFSpec also shows significant performance speedup in the entire RLHF execution.

cs.LG↗

arXiv2025

LOw-cOst yet High-Performant Sparse Matrix-Matrix Multiplication on Arm SME Architectures

Sparse matrix-dense matrix multiplication (SpMM) is a critical kernel in both scientific computing and emerging graph learning workloads. The recent Armv9 architecture introduces Scalable Matrix Extension (SME), enabling tile-based matrix operations with high throughput. However, effectively exploiting both SME and traditional SIMD resources for unstructured sparse workloads remains an open challenge. To address this, we propose LOOPS, a hybrid execution framework that combines row-wise CSR-part with vector-wise BCSR-part layout, enabling cooperative utilization of vector instructions (NEON) and Scalable Matrix Extension (SME) resources. LOOPS supports multi-precision SpMM across FP64, FP32, and FP16 via an adaptive two-level parallelization scheme guided by a lightweight performance model. Experimental results on the entire SuiteSparse on an Apple's M4Pro CPU show that LOOPS achieves average speedups of 9.93$\times$ (FP32)/14.4$\times$ (FP64) against the CPU baseline TACO and 71.3$\times$ (FP32)/54.8$\times$ (FP64) with respect to Armadillo. A comparison of LOOPS running on the same CPU with two GPU methods (cuSPARSE, Magicube) executed on an NVIDIA A100 GPU show average speedups for LOOPS between 19.8$\times$ and 33.5$\times$, depending on the precision. Notably, LOOPS delivers significantly better energy efficiency than the GPU codes on the A100 GPU.

cs.DC↗

arXiv2025

From Geostrophic to Magnetically-Damped Turbulence in Liquid Metal Rotating Magnetoconvection

Understanding planetary core convection dynamics requires the study of convective flows in which the Coriolis and Lorentz forces attain a leading-order, so-called magnetostrophic balance. Experimental investigations of rotating magnetoconvection (RMC) in the magnetostrophic regime are therefore essential to broadly characterize the properties of local-scale planetary core flow. Towards this end, we present here the first thermovelocimetric measurements of magnetostrophic, liquid metal convection, which are made using liquid gallium as the working fluid, at moderate rotation rates (Ekman numbers $10^{-4} \leq Ek\leq 10^{-5}$) and in the presence of dynamically strong magnetic fields (Elsasser number $Λ=1$). Complementary rotating convection (RC) experiments are performed at the same rotation rates to serve as reference cases. Our RMC velocity measurements adequately follow a geostrophic turbulent scaling for cases in which local-scale convective inertial forces exceed the Lorentz forces in the fluid bulk. In cases where Lorentz forces exceed local-scale inertia ($N_\ell \gtrsim 3$), the root-mean-square RMC velocities are magnetically damped, yielding values below the geostrophic turbulent RC scaling prediction. An enhancement in heat transfer is observed, which we attribute to the increased coherence of vertically aligned magnetostrophic convective flow. Extrapolating these laboratory results, we predict that convection-scale flows in Earth's core occur in the magnetically damped $N_\ell \gtrsim 3$ regime with Rayleigh number values between $10^{24}$ and $10^{26}$.

physics.flu-dyn↗

arXiv2025

Diffusivity-Free Turbulence in Liquid Metal Rotating Rayleigh-Bénard Convection Experiments

Convection in planets and stars is predicted to occur in the "ultimate regime'' of diffusivity-free, rapidly rotating turbulence, in which flows are characteristically unaffected by viscous and thermal diffusion. Boundary layer diffusion, however, has historically hindered experimental study of this regime. Here, we utilize the boundary-independent oscillatory thermal-inertial mode of rotating convection to realize the diffusivity-free scaling in liquid metal laboratory experiments. This oscillatory style of convection arises in rotating liquid metals (low Prandtl number fluids) and is driven by the temperature gradient in the fluid bulk, thus remaining independent of diffusive boundary dynamics. We triply verify the existence of the diffusivity-free regime via measurements of heat transfer efficiency $Nu$, dimensionless flow velocities $Re$, and internal temperature anomalies $θ$, all of which are in quantitative agreement with planar asymptotically-reduced models. Achieving the theoretical diffusivity-free scalings in desktop-sized laboratory experiments provides the validation necessary to extrapolate and predict the convective flows in remote geophysical and astrophysical systems.

physics.flu-dyn↗

arXiv2024

Thermovelocimetric Characterization of Liquid Metal Convection in a Rotating Slender Cylinder

Rotating turbulent convection occurs ubiquitously in natural convective systems encompassing planetary cores, oceans, and atmospheres, as well as in many industrial applications. While the global heat and mass transfer of water-like rotating Rayleigh-Bénard convection is well-documented, the characteristics of rotating convection in liquid metals remain less well understood. In this study, we characterize rotating Rayleigh-Bénard convection in liquid gallium (Prandtl number $Pr \approx 0.027$) within a slender cylinder (diameter-to-height aspect ratio $Γ= D/H = 1/2$) using novel thermovelocimetric diagnostic techniques that integrate simultaneous multi-point thermometry and ultrasonic Doppler velocity measurements. This approach experimentally reveals the formation of a stable azimuthal wavenumber $m = 2$ global-scale vortical structure at low supercriticality. We propose that enhanced wall modes facilitated by the slender cylinder geometry interact with the bulk flow to create these large-scale axialized vortices. Our findings extend results from the previous $Pr \sim 1$ studies across various cylindrical aspect ratios. In particular, we find evidence of a different scaling for wall mode precession frequency that possibly exists in liquid metal, offering new insights into the coupling effects in low-$Pr$ rotating convective turbulence.

physics.flu-dyn↗

arXiv2024

FreeMHD: validation and verification of the open-source, multi-domain, multi-phase solver for electrically conductive flows

The extreme heat fluxes in the divertor region of tokamaks may require an alternative to solid plasma-facing components, for the extraction of heat and the protection of the surrounding walls. Flowing liquid metals are proposed as an alternative, but raise additional challenges that require investigation and numerical simulations. Free surface designs are desirable for plasma-facing components (PFCs), but steady flow profiles and surface stability must be ensured to limit undesirable interactions with the plasma. Previous studies have mainly used steady-state, 2D, or simplified models for internal flows and have not been able to adequately model free-surface liquid metal (LM) experiments. Therefore, FreeMHD has been recently developed as an open-source magnetohydrodynamics (MHD) solver for free-surface electrically conductive flows subject to a strong external magnetic field. The FreeMHD solver computes incompressible free-surface flows with multi-region coupling for the investigation of MHD phenomena involving fluid and solid domains. The model utilizes the finite-volume OpenFOAM framework under the low magnetic Reynolds number approximation. FreeMHD is validated using analytical solutions for the velocity profiles of closed channel flows with various Hartmann numbers and wall conductance ratios. Next, experimental measurements are then used to verify FreeMHD, through a series of cases involving dam breaking, 3D magnetic fields, and free-surface LM flows. These results demonstrate that FreeMHD is a reliable tool for the design of LM systems under free surface conditions at the reactor scale. Furthermore, it is flexible, computationally inexpensive, and can be used to solve fully 3D transient MHD flows.

physics.plasm-ph↗

arXiv2024

Magneto-Stokes Flow in a Shallow Free-Surface Annulus

In this study, we analyse "magneto-Stokes" flow, a fundamental magnetohydrodynamic (MHD) flow that shares the cylindrical-annular geometry of the Taylor-Couette cell, but uses applied electromagnetic forces to circulate a free-surface layer of electrolyte at low Reynolds numbers. The first complete, analytical solution for time-dependent magneto-Stokes flow is presented and validated with coupled laboratory and numerical experiments. Three regimes are distinguished (shallow-layer, transitional, and deep-layer flow regimes), and their influence on the efficiency of microscale mixing is clarified. The solution in the shallow-layer limit belongs to a newly-identified class of MHD potential flows, and thus induces mixing without the aid of axial vorticity. We show that these shallow-layer magneto-Stokes flows can still augment mixing in distinct Taylor dispersion and advection-dominated mixing regimes. The existence of enhanced mixing across all three distinguished flow regimes is predicted by asymptotic scaling laws and supported by three-dimensional numerical simulations. Mixing enhancement is initiated with the least electromagnetic forcing in channels with order-unity depth-to-gap-width ratios. If the strength of the electromagnetic forcing is not a constraint, then shallow-layer flows can still yield the shortest mixing times in the advection-dominated limit. Our robust description of momentum evolution and mixing of passive tracers makes the annular magneto-Stokes system fit for use as an MHD reference flow.

physics.flu-dyn↗

arXiv2024

CoNST: Code Generator for Sparse Tensor Networks

Sparse tensor networks are commonly used to represent contractions over sparse tensors. Tensor contractions are higher-order analogs of matrix multiplication. Tensor networks arise commonly in many domains of scientific computing and data science. After a transformation into a tree of binary contractions, the network is implemented as a sequence of individual contractions. Several critical aspects must be considered in the generation of efficient code for a contraction tree, including sparse tensor layout mode order, loop fusion to reduce intermediate tensors, and the interdependence of loop order, mode order, and contraction order. We propose CoNST, a novel approach that considers these factors in an integrated manner using a single formulation. Our approach creates a constraint system that encodes these decisions and their interdependence, while aiming to produce reduced-order intermediate tensors via fusion. The constraint system is solved by the Z3 SMT solver and the result is used to create the desired fused loop structure and tensor mode layouts for the entire contraction tree. This structure is lowered to the IR of the TACO compiler, which is then used to generate executable code. Our experimental evaluation demonstrates very significant (sometimes orders of magnitude) performance improvements over current state-of-the-art sparse tensor compiler/library alternatives.

cs.PL↗

arXiv2023

The Transition from Wall Modes to Multimodality in Liquid Gallium Magnetoconvection

Coupled laboratory-numerical experiments of Rayleigh-Bénard convection (RBC) in liquid gallium subject to a vertical magnetic field are presented. The experiments are carried out in two cylindrical containers with diameter-to-height aspect ratio $Γ= 1.0$ and $2.0$ at varying thermal forcing (Rayleigh numbers $10^5 \lesssim Ra \lesssim 10^8$) and magnetic field strength (Chandrasekhar numbers $0\lesssim Ch \lesssim 3\times 10^5$). Laboratory measurements and numerical simulations confirm that magnetoconvection in our finite cylindrical tanks onsets via non-drifting wall-attached modes, in good agreement with asymptotic predictions for a semi-infinite domain. With increasing supercriticality, the experimental and numerical thermal measurements and the numerical velocity data reveal transitions between wall mode states with different azimuthal mode numbers and between wall-dominated convection to wall and interior multimodality. These transitions are also reflected in the heat transfer data, which combined with previous studies, connect onset to supercritical turbulent behaviors in liquid metal magnetoconvection over a large parameter space. The gross heat transfer behaviors between magnetoconvection and rotating convection in liquid metals are compared and discussed.

physics.flu-dyn↗

arXiv2021

Thermoelectric Precession in Turbulent Magnetoconvection

We present laboratory measurements of the interaction between thermoelectric currents and turbulent magnetoconvection. In a cylindrical volume of liquid gallium heated from below and cooled from above and subject to a vertical magnetic field, it is found that the large scale circulation (LSC) can undergo a slow axial precession. Our experiments demonstrate that this LSC precession occurs only when electrically conducting boundary conditions are employed, and that the precession direction reverses when the axial magnetic field direction is flipped. A thermoelectric magnetoconvection (TEMC) model is developed that successfully predicts the zeroth-order magnetoprecession dynamics. Our TEMC magnetoprecession model hinges on thermoelectric current loops at the top and bottom boundaries, which create Lorentz forces that generate horizontal torques on the overturning large-scale circulatory flow. The thermoelectric torques in our model act to drive a precessional motion of the LSC. This model yields precession frequency predictions that are in good agreement with the experimental observations. We postulate that thermoelectric effects in convective flows, long argued to be relevant in liquid metal heat transfer and mixing processes, may also have applications in planetary interior magnetohydrodynamics.

physics.flu-dyn↗

arXiv2021

Dependence-Aware, Unbounded Sound Predictive Race Detection

Data races are a real problem for parallel software, yet hard to detect. Sound predictive analysis observes a program execution and detects data races that exist in some other, unobserved execution. However, existing predictive analyses miss races because they do not scale to full program executions or do not precisely incorporate data and control dependence. This paper introduces two novel, sound predictive approaches that incorporate data and control dependence and handle full program executions. An evaluation using real, large Java programs shows that these approaches detect more data races than the closest related approaches, thus advancing the state of the art in sound predictive race detection.

cs.PL↗

arXiv2021

Efficient distributed algorithms for Convolutional Neural Networks

Several efficient distributed algorithms have been developed for matrix-matrix multiplication: the 3D algorithm, the 2D SUMMA algorithm, and the 2.5D algorithm. Each of these algorithms was independently conceived and they trade-off memory needed per node and the inter-node data communication volume. The convolutional neural network (CNN) computation may be viewed as a generalization of matrix-multiplication combined with neighborhood stencil computations. We develop communication-efficient distributed-memory algorithms for CNNs that are analogous to the 2D/2.5D/3D algorithms for matrix-matrix multiplication.

cs.DC↗

arXiv2021

Deformable and robust core-shell protein microcapsules templated by liquid-liquid phase separated microdroplets

Microcapsules are a key class of microscale materials with applications in areas ranging from personal care to biomedicine, and with increasing potential to act as extracellular matrix (ECM) models of hollow organs or tissues. Such capsules are conventionally generated from non-ECM materials including synthetic polymers. Here, we fabricated robust microcapsules with controllable shell thickness from physically- and enzymatically-crosslinked gelatin and achieved a core-shell architecture by exploiting a liquid-liquid phase separated aqueous dispersed phase system in a one-step microfluidic process. Microfluidic mechanical testing revealed that the mechanical robustness of thicker-shell capsules could be controlled through modulation of the shell thickness. Furthermore, the microcapsules demonstrated environmentally-responsive deformation, including buckling by osmosis and external mechanical forces. A sequential release of cargo species was obtained through the degradation of the capsules. Stability measurements showed the capsules were stable at 37 °C for more than two weeks. Finally, all-aqueous liquid-liquid phase separated and multiphase liquid-liquid phase separated systems were generated with the gel-sol transition of microgel precursors. These smart capsules are promising models of hollow biostructures, microscale drug carriers, and building blocks or compartments for active soft materials and robots.

cond-mat.soft↗

arXiv2021

Analytical Characterization and Design Space Exploration for Optimization of CNNs

Moving data through the memory hierarchy is a fundamental bottleneck that can limit the performance of core algorithms of machine learning, such as convolutional neural networks (CNNs). Loop-level optimization, including loop tiling and loop permutation, are fundamental transformations to reduce data movement. However, the search space for finding the best loop-level optimization configuration is explosively large. This paper develops an analytical modeling approach for finding the best loop-level optimization configuration for CNNs on multi-core CPUs. Experimental evaluation shows that this approach achieves comparable or better performance than state-of-the-art libraries and auto-tuning based optimizers for CNNs.

cs.LG↗