Northwestern Polytechnic University

Rectified Flow (RF) has been widely used as an effective generative model. Although RF is primarily based on probability flow Ordinary Differential Equations (ODE), recent studies have shown that injecting noise through reverse-time Stochastic Differential Equations (SDE) for sampling can achieve superior generative performance. Inspired by Positive-incentive Noise (Pi-noise), we propose an innovative generative algorithm to train Pi-noise generators, namely Rectified Noise (RN), which improves the generative performance by injecting Pi-noise into the velocity field of pre-trained RF models. After introducing the Rectified Noise pipeline, pre-trained RF models can be efficiently transformed into Pi-noise generators. We validate Rectified Noise by conducting extensive experiments across various model architectures on different datasets. Notably, we find that: (1) RF models using Rectified Noise reduce FID from10.16 to 9.05 on ImageNet-1k. (2) The models of Pi-noise generators achieve improved performance with only 0.39% additional training parameters.

Vision-and-Language Navigation (VLN) plays a critical role in tasks of embodied AI, particularly in unseen environments following natural language instructions. Recent advancements leverage large language models (LLMs) to improve the accuracy and generalizability of VLN systems by encoding image sequences as dense token representations. However, this tokenization approach incurs substantial computational overhead due to two key inefficiencies: 1) ego-centric camera views often include navigation-irrelevant re- gions (e.g., sky or distant backgrounds), and 2) high-frame-rate image sequences introduce temporal redundancy. To address these challenges, we propose Spatial-Temporal Efficient Visual Token Pruning (STEP-Nav), a unified frame- work that simultaneously prunes redundant visual tokens and fine-tunes VLN models to preserve navigation performance. In particular, STEP-Nav incorporates a distance- and content-aware token evaluation mechanism to remove irrelevant tokens at the spatial level, along with temporal level similarity-based filtering to reduce redundancy across sequential frames. To ensure pruning does not harm task performance, we introduce a distortion-aware fine-tuning strategy that aligns pruned-token representations with their full-token counterparts while maintaining navigation accuracy. Experiments on the R2R and RxR benchmarks using Navid-CE and NavGPT-2 as base models demonstrate that STEP-Nav preserves over 95% of the performance while reducing 66.7% of tokens, outperforming existing token pruning baselines.

Aircraft interacts with the ground through landing gear system, while taxiing down the runway.Under rain or snow conditions, water film splashed by aircraft tires during taxiing on water-covered runways adheres to the landing gear, leading to ice accretion under low-temperature conditions, which may directly compromise aircraft takeoff and landing safety. Due to the substantial resource consumption required for experimental investigation of this problem, numerical simulation was employed to model the phenomenon.The water film icing process can be divided into two main stages: the formation of water film on the landing gear caused by tire spray, and the subsequent icing and heat transfer process of the splashed water film on the landing gear, this study conservatively simplifies the tire spray phenomenon as a typical liquid-solid two-phase coupling problem. A Finite Volume Model simulating water accumulation dynamics on landing gear tires was developed in Fluent software, based on extreme runway water accumulation conditions and extreme navigation weather scenarios during takeoff of a domestic civil aircraft model. The model incorporates coupled tire-water film-landing gear interactions. The VOF (Volume of Fluid) model and mesh motion are employed to calculate the water film thickness on the landing gear. A conservative thermodynamic equilibrium equation is then constructed based on boundary conditions to estimate the ice accretion thickness. The article analyzed the influence of external factors (aerodynamic drag, gravitational force, etc.) on water film thickness. The calculation results indicate that a conservative estimation of ice accretion derived from water film thickness does not compromise takeoff and landing safety, thereby validating that under rainy or snowy weather conditions with water accumulation on the runway, the conservative maximum icing thickness resulting from wheel spray water and landing gear water accumulation during aircraft taxiing will not exceed a level sufficient to pose a threat to aircraft takeoff and landing safety.

The pathological process of type 2 diabetes mellitus (T2DM) is closely associated with chronic low−grade inflammation. High mobility group box 1 (HMGB1), a key damage−associated molecular pattern (DAMP), is frequently dysregulated in T2DM and is implicated in promoting insulin resistance (IR), β cell dysfunction, and the progression of multiple complications—including cardiovascular disease, nephropathy, cognitive impairment, myopathy, and dyslipidemia—primarily through activating signaling pathways such as RAGE/TLR4−NF−κB. Exercise, a cornerstone non−pharmacological intervention, effectively mitigates HMGB1−driven pathology through multifaceted mechanisms. These include direct downregulation of HMGB1 expression and suppression of its downstream inflammatory pathways, as well as indirect effects via improved glycemic control, enhancing autophagy, and reduced oxidative stress. This review aims to systematically examine the evidence for the role of HMGB1 in T2DM pathogenesis and its complications, and to evaluate exercise as a potential strategy to target this inflammatory pathway, thereby providing a theoretical framework for future therapeutic approaches.

Hafnia-based ferroelectric materials have garnered considerable attention due to robust ferroelectricity in ultrathin films and excellent compatibility with silicon-based technology. Theoretical predictions of the polarization along [001] direction of ferroelectric HfO2 are 50 μC/cm2 and 70 μC/cm2, respectively, depending on the switching mechanism. However, most experimental observations of the intrinsic polarization are much lower than these predictions. Here, we report that an intrinsic remnant polarization up to 40 μC/cm2 is achieved in epitaxially grown (111)-oriented Hf0.5Zr0.5O2/Hf0.9La0.1O2 multilayer film, corresponding to 69.3 μC/cm2 along [001], approaching the theoretical limit. Structural analyses reveal a rhombohedral-distorted orthorhombic phase in the Hf0.5Zr0.5O2/Hf0.9La0.1O2 multilayers, stabilized by an in-plane compressive strain. Density functional theory calculations demonstrate that La doping in Hf0.5Zr0.5O2/Hf0.9La0.1O2 promotes an unconventional switching pathway and contributes to the high intrinsic polarization. These findings provide a compelling strategy for achieving high intrinsic polarization and establish a design paradigm for high-performance hafnia-based ferroelectric devices.

Numerical simulation of ternary flow is still a challenging problem due to the presence of complex interfacial dynamics and irregular geometric domains. Conventional lattice Boltzmann methods (LBM) are limited by their dependence on structured grids and the strict coupling between spatial and temporal discretizations, which reduces their flexibility in handling such complex systems. This paper presents a novel generalized semi-Lagrangian meshfree lattice Boltzmann method (SL-M-LBM) for ternary fluid flows by using a two-component phase-field model. Our approach fundamentally decouples the discretization by integrating a semi-Lagrangian streaming algorithm with moving least squares (MLS) reconstruction. This enables entirely mesh-free simulations with flexible, non-uniform node distributions and independent control of time stepping, thereby enhancing numerical stability and allowing for local refinement in complex domains. To solve the hydrodynamic equations for incompressible flows and numerically capture interface evolution, our approach creates a Lattice Boltzmann (LB) model with two LB equations (LBEs) for phase-field evolution and a single LB equation (LBE) for hydrodynamics. A wide range of benchmark tests are conducted to evaluate the model, such as Zalesak's rotating disk, diagonal translation, two circular interfaces deformation under shear flow, binary and ternary Rayleigh–Taylor instability, compound droplet passing through a capillary throat under pressure-driven, interaction between a droplet and a rising bubble under gravity, and finally liquid lens spreading. The simulation results indicate that the SL-M-LBM performs effectively for many flow conditions with accurate interface tracking and good mass conservation ability even with non-uniform node distributions. Furthermore, the model can handle irregular geometric domains, high density ratios, and highlights its potential for versatile and efficient simulation of complex multiphase systems containing three immiscible fluids.