A5016703530- Video Surveillance and Tracking Methods
- Machine Learning and ELM
- Domain Adaptation and Few-Shot Learning
- Face and Expression Recognition
- Neural Networks and Applications
- Sparse and Compressive Sensing Techniques
- Enhanced Oil Recovery Techniques
- Hydrocarbon exploration and reservoir analysis
- Advanced Neural Network Applications
- Adversarial Robustness in Machine Learning
- Image Enhancement Techniques
- Petroleum Processing and Analysis
- Gaussian Processes and Bayesian Inference
- Advanced Vision and Imaging
- Human Pose and Action Recognition
- Stochastic Gradient Optimization Techniques
- Polymer crystallization and properties
- Generative Adversarial Networks and Image Synthesis
- Advanced Control Systems Design
- Model Reduction and Neural Networks
- Machine Learning and Data Classification
- Polymer Foaming and Composites
- Injection Molding Process and Properties
- Polymer Surface Interaction Studies
- Control Systems and Identification
University of Warwick
2024
École Polytechnique Fédérale de Lausanne
2022
Sinopec (China)
2022
Chongqing University of Posts and Telecommunications
2022
KU Leuven
2018-2021
Chinese Academy of Sciences
2015-2020
Institute of Chemistry
2015-2020
Liaoning University
2017-2020
University of Chinese Academy of Sciences
2017-2020
Shanghai Jiao Tong University
2015-2019
The class of random features is one the most popular techniques to speed up kernel methods in large-scale problems. Related works have been recognized by NeurIPS Test-of-Time award 2017 and ICML Best Paper Finalist 2019. body work on has grown rapidly, hence it desirable a comprehensive overview this topic explaining connections among various algorithms theoretical results. In survey, we systematically review from past ten years. First, motivations, characteristics contributions...
Abstract The morphological feature of microparts evolved during micro‐injection molding may differ from that the macroparts prepared by conventional injection molding, resulting in specific physical properties. In this study, isotactic polypropylene (iPP) with 200 µm thickness and 2000 were prepared, their comparison was investigated means polarized light microscopy (PLM), scanning electron (SEM), differential calorimeter (DSC), wide‐angle X‐ray diffraction (WAXD). results presented some...
This paper studies how to improve the performance of Low-Rank Adaption (LoRA) as guided by our theoretical analysis. Our first set results show that for random initialization and linear models, \textit{i)} LoRA will align certain singular subspace one-step gradient full fine-tuning; \textit{ii)} preconditioners convergence in high-rank case. These insights motivate us focus on preconditioned using a specific spectral strategy aligning with subspaces. For both nonlinear we prove alignment...
We investigate double descent and scaling laws in terms of weights rather than the number parameters. Specifically, we analyze linear random features models using deterministic equivalence approach from matrix theory. precisely characterize how norm concentrate around quantities elucidate relationship between expected test error norm-based capacity (complexity). Our results rigorously answer whether exists under reshape corresponding laws. Moreover, they prompt a rethinking data-parameter...
Desorption of asphaltenes from silica-coated quartz crystals upon exposure to a series saline solutions was studied through the measurements crystal microbalance with dissipation (QCM-D), atomic force microscopy (AFM), and contact angle. Interestingly, it found that mass loading thickness asphaltene film decreased during injection sodium chloride solution at concentrations ranging 1 10 mM, surface tending be hydrophilic, whereas increased gradually when concentration 1000 inclined...
In this paper, we propose a novel and robust tracking framework based on online discriminative low-rank dictionary learning. The primary aim of paper is to obtain compact dictionaries that can provide good representations both target background. We accomplish by exploiting the recovery ability matrices. That if assume data from same class are linearly correlated, then corresponding basis vectors learned training set each shall render become approximately low-rank. proposed learning technique...
In this paper, we propose a novel matching based tracker by investigating the relationship between template and recent popular correlation filter trackers (CFTs). Compared to operation in CFTs, sophisticated similarity metric termed mutual buddies is proposed exploit of multiple reciprocal nearest neighbors for target matching. By doing so, our obtains powerful discriminative ability on distinguishing background as demonstrated both empirical theoretical analyses. Besides, instead utilizing...
The presence of thin aqueous films and their stability have a profound effect on the interactions between oil/brine/rock interfaces. In previous report, we proposed that hydration forces, originating from overlap hydrated layers different surfaces in sodium chloride, played an important role at short range. present work, divalent ions were introduced to liquid and, mechanisms improving oil recovery low-salinity brine molecular level revealed. Through direct force-measuring technique chemical...
The adhesion of mussel foot proteins (Mfps) to a variety surfaces has been widely investigated, but the mechanisms behind with different properties are far from being understood.
Visual tracking is complicated due to factors, such as occlusion, background clutter, abrupt target motion, and illumination variations, among others. In recent years, subspace representation sparse coding techniques have demonstrated significant improvements in tracking. However, performance gain has been at the expense of losing locality similarity attributes instances be encoded. this paper, a graph regularized locality-constrained (GRLC) technique that encapsulates local manifold...
Thermodynamic phase behavior is affected by curved interfaces in micro- and nanoscale systems. For example, capillary freezing point depression associated with the pressure difference between solid liquid phases caused interface curvature. In this study, thermal, mechanical, chemical equilibrium conditions are derived for binary solid–liquid a due to confinement capillary. This derivation shows equivalence of most general forms Gibbs–Thomson Ostwald–Freundlich equations. As an effect...
Abstract The petroleum industry has focused on the modification of crude oil composites to decrease their adhesion onto materials and thus facilitate extraction, transportation, storage, processing. However, these methods such as heating, dilution, emulsification, or additives are often accompanied by significant costs suffer from various limitations. Herein, we present a conceptually different coating strategy that allows many substrates repel oil. novel was achieved via fully waterborne...
Random features is one of the most popular techniques to speed up kernel methods in large-scale problems. Related works have been recognized by NeurIPS Test-of-Time award 2017 and ICML Best Paper Finalist 2019. The body work on random has grown rapidly, hence it desirable a comprehensive overview this topic explaining connections among various algorithms theoretical results. In survey, we systematically review from past ten years. First, motivations, characteristics contributions...
In this study, Kalman filters for continuous‐time linear fractional‐order systems are studied with coloured process and measurement noise, respectively. By average derivative, or noise discretised. To deal noises, the authors construct an augmented system respect to state, noise. Furthermore, filter using derivative is proposed. This improves accuracy of state estimation filtering effect noises. Finally, they give two examples verify correctness validity proposed algorithm.
In kernel methods, the kernels are often required to be positive definitethat restricts use of many indefinite kernels. To consider those nonpositive definite kernels, in this paper, we aim build an learning framework for logistic regression (KLR). The proposed KLR (IKLR) model is analyzed reproducing Kreĭn spaces and then becomes nonconvex. Using decomposition a kernel, derived IKLR can decomposed into difference two convex functions. Accordingly, concave-convex procedure (CCCP) introduced...