- Recommender Systems and Techniques
- Topic Modeling
- Advanced Graph Neural Networks
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Advanced Neural Network Applications
- Medical Imaging and Analysis
- Advanced Bandit Algorithms Research
- Microwave Imaging and Scattering Analysis
- Advanced optical system design
- Indoor and Outdoor Localization Technologies
- Bone Tissue Engineering Materials
- Energy Load and Power Forecasting
- Service-Oriented Architecture and Web Services
- Radiomics and Machine Learning in Medical Imaging
- Spinal Fractures and Fixation Techniques
- Image and Video Quality Assessment
- Text and Document Classification Technologies
- Multimodal Machine Learning Applications
- Consumer Market Behavior and Pricing
- Advanced Decision-Making Techniques
- Non-Destructive Testing Techniques
- Photonic and Optical Devices
- Sentiment Analysis and Opinion Mining
- Web Data Mining and Analysis
Huawei Technologies (China)
2012-2025
Inner Mongolia Electric Power (China)
2022-2025
Kunming Medical University
2022-2025
Shanghai Jiao Tong University
2012-2024
Ruijin Hospital
2019-2023
Sichuan University
2022-2023
West China Hospital of Sichuan University
2023
University of Illinois Urbana-Champaign
2018-2023
Jiangsu Frontier Electric Technology Co., Ltd. (China)
2022
Western University
2019-2022
We present a class of efficient models called MobileNets for mobile and embedded vision applications. are based on streamlined architecture that uses depth-wise separable convolutions to build light weight deep neural networks. introduce two simple global hyper-parameters efficiently trade off between latency accuracy. These allow the model builder choose right sized their application constraints problem. extensive experiments resource accuracy tradeoffs show strong performance compared...
Inverted bottleneck layers, which are built upon depth-wise convolutions, have been the predominant building blocks in state-of-the-art object detection models on mobile devices. In this work, we investigate optimality of design pattern over a broad range accelerators by revisiting usefulness regular convolutions. We discover that convolutions potent component to boost latency-accuracy trade-off for accelerators, provided they placed strategically network via neural architecture search. By...
With the rapid development of online services and web applications, recommender systems (RS) have become increasingly indispensable for mitigating information overload matching users’ needs by providing personalized suggestions over items. Although RS research community has made remarkable progress past decades, conventional recommendation models (CRM) still some limitations, e.g. , lacking open-domain world knowledge, difficulties in comprehending underlying preferences motivations....
Image re-ranking is effective in improving performance of text-based image searches. However, improvements from existing algorithms are limited by two factors: one that the associated textual information images often mismatches their actual visual contents; other a visual's features cannot accurately describe semantic similarities between images. In this paper, we adopt click data to bridge gap. We propose novel multi-view hypergraph-based learning (MHL) method adaptively integrates with...
Quantifying full left ventricular (LV) metrics including cavity area, myocardium dimensions and wall thicknesses from cardiac magnetic resonance (MR) images, then assessing regional global function plays a crucial role in clinical practice. However, due to highly variable structures across different subjects, it is challenging obtain an accurate estimation of LV metrics. In this paper, we propose novel deep learning framework, called cascaded segmentation regression network (CSRNet), improve...
Factorization Machine (FM) is an effective solution for context-aware recommender systems (CARS) which models second-order feature interactions by inner product. However, it insufficient to capture high-order and nonlinear interaction signals. While several recent efforts have enhanced FM with neural networks, they assume the embedding dimensions are independent from each other model in a rather implicit manner. In this paper, we propose Convolutional (CFM) address above limitations....
Click-Through Rate (CTR) prediction is critical for industrial recommender systems, where most deep CTR models follow an Embedding & Feature Interaction paradigm. However, the majority of methods focus on designing network architectures to better capture feature interactions while embedding, especially numerical features, has been overlooked. Existing approaches features are difficult informative knowledge because low capacity or hard discretization based offline expertise engineering. In...
Sequential recommendation holds the promise of understanding user preference by capturing successive behavior correlations. Existing research focus on designing different models for better fitting offline datasets. However, observational data may have been contaminated exposure or selection biases, which renders learned sequential unreliable. In order to solve this fundamental problem, in paper, we propose reformulate task with potential outcome framework, where are able clearly understand...
With large language models (LLMs) achieving remarkable breakthroughs in NLP domains, LLM-enhanced recommender systems have received much attention and been actively explored currently. In this paper, we focus on adapting empowering a pure model for zero-shot few-shot recommendation tasks. First foremost, identify formulate the lifelong sequential behavior incomprehension problem LLMs i.e., fail to extract useful information from textual context of long user sequence, even if length is far...
Click-through rate (CTR) prediction has become increasingly indispensable for various Internet applications. Traditional CTR models convert the multi-field categorical data into ID features via one-hot encoding, and extract collaborative signals among features. Such a paradigm suffers from problem of semantic information loss. Another line research explores potential pretrained language (PLMs) by converting input textual sentences through hard prompt templates. Although are preserved, they...
Recommender system plays a vital role in various online services. However, its insulated nature of training and deploying separately within specific closed domain limits access to open-world knowledge. Recently, the emergence large language models (LLMs) has shown promise bridging this gap by encoding extensive world knowledge demonstrating reasoning capabilities. Nevertheless, previous attempts directly use LLMs as recommenders cannot meet inference latency demand industrial recommender...
Estimation of the basic parameters, wall thickness and dielectric constant, is important in through-the-wall radar imaging. Ambiguities characteristics will degrade image focusing quality synthetic-aperture radar. In order to obtain a quick precise estimation an equivalent propagation model electromagnetic wave air-wall-air medium first developed this paper. According model, two filter-based approaches, denoted respectively as echo-domain-filter-based method image-domain-filter-based method,...
We implement a passive remote keystroke detection mechanism using only changes in the wireless channel. The algorithm does not require user to wear any active devices nor it change user's transmission technique. receiver system is implemented with five antennas. cancel signals received on multiple key insight realizing fine-grained localization exploit extremely high sensitivity of cancellation performance (interference full-duplex for example) exact amplitude and phase matching. design...
Luminescence nanomaterial-based lateral flow assay (LFA) is promising for point-of-care tests. However, the detection sensitivity and accuracy are often affected by interferences of autofluorescence photon scattering from nitrocellulose membrane colored plasma. Here, we describe a near-infrared to upconversion nanoparticle (UCNP) immunolabeled LFA background-free chromatographic sepsis biomarker procalcitonin (PCT) in clinical human This immunolabeling enables both light excitation (at ∼980...
The formation of a calcified cartilaginous callus (CACC) is crucial during bone repair. CACC can stimulate the invasion type H vessels into to couple angiogenesis and osteogenesis, induce osteoclastogenesis resorb matrix, promote osteoclast secretion factors enhance ultimately achieving replacement cartilage with bone. In this study, porous polycaprolactone/hydroxyapatite-iminodiacetic acid-deferoxamine (PCL/HA-SF-DFO) 3D biomimetic developed using printing. structure mimic pores formed by...
With the explosive growth of commercial applications recommender systems, multi-scenario recommendation (MSR) has attracted considerable attention, which utilizes data from multiple domains to improve their performance simultaneously. However, training a unified deep system (DRS) may not explicitly comprehend commonality and difference among domains, whereas an individual model for each domain neglects global information incurs high computation costs. Likewise, fine-tuning on is inefficient,...
Multi-Domain Recommendation (MDR) has gained significant attention in recent years, which leverages data from multiple domains to enhance their performance concurrently. However, current MDR models are confronted with two limitations. Firstly, the majority of these adopt an approach that explicitly shares parameters between domains, leading mutual interference among them. Secondly, due distribution differences utilization static existing methods limits flexibility adapt diverse domains. To...
Multi-types of user behavior data (e.g., clicking, adding to cart, and purchasing) are recorded in most real-world recommendation scenarios, which can help learn users' multi-faceted preferences. However, it is challenging explore multi-behavior due the unbalanced distribution sparse target behavior, lead inadequate modeling high-order relations when treating "as features" gradient conflict multi-task learning labels". In this paper, we propose CIGF, a Compressed Interaction Graph based...
With the explosive growth of various commercial scenarios, there is an increasing number studies on multi-scenario recommendation (MSR) which trains recommender system with data from multiple aiming to improve performance all these scenarios synchronously. However, due large discrepancy in interactions among domains, models usually suffer insufficient learning and negative transfer especially cold-start thus exacerbating sparsity issue. To fill this gap, work we propose a novel diffusion...
With the rapid expansion of digital music formats, it's indispensable to recommend users with their favorite music. For recommendation, users' personality and emotion greatly affect preference, respectively in a long-term short-term manner, while rich social media data provides effective feedback on these information. In this paper, aiming at recommendation platforms, we propose Personality Emotion Integrated Attentive model (PEIA), which fully utilizes comprehensively taste (personality)...