A5101846033- Advanced Neural Network Applications
- Advanced Computational Techniques and Applications
- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Advanced Vision and Imaging
- Advanced Image Processing Techniques
- Network Security and Intrusion Detection
- Anomaly Detection Techniques and Applications
- Data Mining Algorithms and Applications
- Video Analysis and Summarization
- Antimicrobial Peptides and Activities
- COVID-19 diagnosis using AI
- Video Surveillance and Tracking Methods
- Advanced Decision-Making Techniques
- Time Series Analysis and Forecasting
- Target Tracking and Data Fusion in Sensor Networks
- Machine Learning and Algorithms
- Recommender Systems and Techniques
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Traffic Prediction and Management Techniques
- Automated Road and Building Extraction
- Complex Systems and Time Series Analysis
- Robotics and Sensor-Based Localization
- Biochemical and Structural Characterization
Jilin University
2024
Harbin Institute of Technology
2021-2024
ITMO University
2019-2024
Baoding University
2024
Jilin Province Science and Technology Department
2024
Georgia Institute of Technology
2021-2023
Sungkyunkwan University
2023
HeidelbergCement (United States)
2022
Sensors (United States)
2021
Georgia Tech Research Institute
2021
Reliable and accurate 3D object detection is a necessity for safe autonomous driving. Although LiDAR sensors can provide point cloud estimates of the environment, they are also prohibitively expensive many settings. Recently, introduction pseudo-LiDAR (PL) has led to drastic reduction in accuracy gap between methods based on those cheap stereo cameras. PL combines state-of-the-art deep neural networks depth estimation with by converting 2D map outputs inputs. However, so far these two have...
Popular network pruning algorithms reduce redundant information by optimizing hand-crafted models, and may cause suboptimal performance long time in selecting filters. We innovatively introduce adaptive exemplar filters to simplify the algorithm design, resulting an automatic efficient approach called EPruner. Inspired face recognition community, we use a message-passing Affinity Propagation on weight matrices obtain number of exemplars, which then act as preserved EPruner breaks dependence...
Existing online knowledge distillation approaches either adopt the student with best performance or construct an ensemble model for better holistic performance. However, former strategy ignores other students' information, while latter increases computational complexity during deployment. In this article, we propose a novel method distillation, termed feature fusion and self-distillation (FFSD), which comprises two key components: FFSD, toward solving above problems in unified framework....
Abstract The application of deep learning in high‐precision ionospheric parameter prediction has become one the focus space weather research. In this study, an improved model called Mixed Convolutional Neural Networks (CNN)—Bi‐Long Short Term Memory is proposed for predicting future Total Electron Content (TEC). trained using longest available (25 years) Global Ionospheric Maps‐TEC and evaluated accuracy storm predictions. results indicate that historical TEC solar‐geographical reference...
In this paper, we address the problem of monocular depth estimation when only a limited number training image-depth pairs are available. To achieve high regression accuracy, state-of-the-art methods rely on CNNs trained with large pairs, which prohibitively costly or even infeasible to acquire. Aiming break curse such expensive data collections, propose semi-supervised adversarial learning framework that utilizes small in conjunction easily-available images performance. particular, use one...
Generative Adversarial Networks (GANs) have been widely-used in image translation, but their high computational and storage costs impede the deployment on mobile devices. Prevalent methods for CNN compression cannot be directly applied to GANs due specificity of GAN tasks unstable adversarial training. To solve these, this paper, we introduce a novel method, termed DMAD, by proposing Differentiable Mask co-Attention Distillation. The former searches light-weight generator architecture...
Multi-source Domain Adaptation (MDA) aims to transfer knowledge from multiple labeled source domains an unlabeled target domain. Nevertheless, traditional methods primarily focus on achieving inter-domain alignment through sample-level constraints, such as Maximum Mean Discrepancy (MMD), neglecting three pivotal aspects: 1) the potential of data augmentation, 2) significance intra-domain alignment, and 3) design cluster-level constraints. In this paper, we introduce a novel hardness-driven...
Cloud services typically compose of multiple distributed software components that communicate with each other through web service interfaces in the cloud environments. During their long time running, accumulation internal errors or large consumption computing resources will very likely lead to aging problems. In order solve this problem, rejuvenation technology is proposed prevent them from causing more serious failures by restarting running. research field and for services, how accurately...
Lianhui Qin, Lemao Liu, Wei Bi, Yan Wang, Xiaojiang Zhiting Hu, Hai Zhao, Shuming Shi. Proceedings of the 56th Annual Meeting Association for Computational Linguistics (Volume 2: Short Papers). 2018.
We developed a minimum gradient based method to track ridge features in 2D image plot, which is typical data representation many momentum resolved spectroscopy experiments. Through both analytic formulation and numerical simulation, we compare this new with existing DC (distribution curve) higher order derivative analyses. find that the has good noise resilience enhanced contrast especially for weak intensity features, meanwhile preserves quantitative local maxima information from raw image....
Automated machine learning (AutoML) has achieved remarkable progress on various tasks, which is attributed to its minimal involvement of manual feature and model designs. However, most existing AutoML pipelines only touch parts the full pipeline, e.g., neural architecture search or optimizer selection. This leaves potentially important components such as data cleaning ensemble out optimization, still results in considerable human suboptimal performance. The main challenges lie huge space...
In order to improve the precision of network security situation prediction, a combined method for prediction was proposed. For nonlinearity value, Method predicted by using ARMA model and Markov model. On this basis, it optimized results with strategy. The analysis example indicates that can effectively compared single Method.
Recent works have validated the benefit of integrating spatial information into deep networks to improve pixel-level prediction tasks such as monocular depth estimation. However, how efficiently and robustly integrate cues retains an open problem. In this paper, we introduce Side Prediction Aggregation (termed SPA) method enhance embedding scene structural from low-level high-level layers. To estimation accuracy, proposed is further equipped with continuous Spatial Refinement Loss SRL) at...
We present a data association method for vision-based multiple pedestrian tracking, using deep convolutional features to distinguish between different people based on their appearances. These re-identification (re-ID) are learned such that they invariant transformations as rotation, translation, and changes in the background, allowing consistent identification of moving through scene. incorporate re-ID into general likelihood model person experimentally validate this by it perform tracking...
Online Multi-Object Tracking (MOT) has wide applications in time-critical video analysis scenarios, such as robot navigation1 and autonomous driving2. In tracking by-detection, a major challenge of online MOT is how to robustly associate noisy object detections on new frame with previously tracked objects. This paper aims build technology that can track movement people via surveillance cameras are located stores, but not only (theoretically, the algorithm may be applicable location camera at...
Text-to-face (T2F) generation is an emerging research hot spot in multimedia, and its main challenge lies the high fidelity requirement of generated portraits. Many existing works resort to exploring latent space a pre-trained generator, e.g., StyleGAN, which has obvious shortcomings efficiency generalization ability. In this paper, we propose generative network for open-ended text-to-face generation, termed OpenFaceGAN. Differing from StyleGAN-based methods, OpenFaceGAN constructs effective...