A5069418988- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Face recognition and analysis
- Generative Adversarial Networks and Image Synthesis
- COVID-19 diagnosis using AI
- Biometric Identification and Security
- Domain Adaptation and Few-Shot Learning
- Topic Modeling
- Video Surveillance and Tracking Methods
- Multimodal Machine Learning Applications
- Anomaly Detection Techniques and Applications
- Gait Recognition and Analysis
- Direction-of-Arrival Estimation Techniques
- Face and Expression Recognition
- Imbalanced Data Classification Techniques
- Rough Sets and Fuzzy Logic
- Seismic Performance and Analysis
- Adversarial Robustness in Machine Learning
- AI in cancer detection
- Data Mining Algorithms and Applications
- Traffic Prediction and Management Techniques
- Advanced Optical Sensing Technologies
- Natural Language Processing Techniques
- Robotics and Automated Systems
- Autonomous Vehicle Technology and Safety
New York University
2024
Northeastern University
2020-2024
Alibaba Group (Cayman Islands)
2023
Universidad del Noreste
2021
Anhui University of Finance and Economics
2021
Southwest University
2019
National University of Defense Technology
2017
In the background of information age, importance data resources can be imagined, and use means—data mining has also emerged. current situation, all industries are in a relatively equal stage, should make good resources, Apriori algorithm to mine association rules, formulate marketing strategies, promote sales growth slow down loss national GDP. Countries predictions support decisions. Briefly describe basic concepts rules. Only with understanding rules we better understand algorithm, why it...
Vision Transformers (ViTs) have demonstrated powerful representation ability in various visual tasks thanks to their intrinsic data-hungry nature. However, we unexpectedly find that ViTs perform vulnerably when applied face recognition (FR) scenarios with extremely large datasets. We investigate the reasons for this phenomenon and discover existing data augmentation approach hard sample mining strategy are incompatible ViTs-based FR backbone due lack of tailored consideration on preserving...
Semi-Supervised Semantic Segmentation (S4) aims to train a segmentation model with limited labeled images and substantial volume of unlabeled images. To improve the robustness representations, powerful methods introduce pixel-wise contrastive learning approach in latent space (i.e., representation space) that aggregates representations their prototypes fully supervised manner. However, previous contrastive-based S4 merely rely on supervision from model's output (logits) logit during...
Scene completion and forecasting are two popular perception problems in research for mobile agents like autonomous vehicles. Existing approaches treat the isolation, resulting a separate of aspects. In this paper, we introduce novel LiDAR task Occupancy Completion Forecasting (OCF) context driving to unify these aspects into cohesive framework. This requires new algorithms address three challenges altogether: (1) sparse-to-dense reconstruction, (2) partial-to-complete hallucination, (3)...
Tremendous breakthroughs have been developed in Semi-Supervised Semantic Segmentation (S4) through contrastive learning. However, due to limited annotations, the guidance on unlabeled images is generated by model itself, which inevitably exists noise and disturbs unsupervised training process. To address this issue, we propose a robust contrastive-based S4 framework, termed Probabilistic Representation Contrastive Learning (PRCL) framework enhance robustness of We pixel-wise representation...
In the field of human-centric personalized image generation, adapter-based method obtains ability to customize and generate portraits by text-to-image training on facial data. This allows for identity-preserved personalization without additional fine-tuning in inference. Although there are improvements efficiency fidelity, is often a significant performance decrease test following ability, controllability, diversity generated faces compared base model. this paper, we analyze that degradation...
Vision Transformers (ViTs) have demonstrated powerful representation ability in various visual tasks thanks to their intrinsic data-hungry nature. However, we unexpectedly find that ViTs perform vulnerably when applied face recognition (FR) scenarios with extremely large datasets. We investigate the reasons for this phenomenon and discover existing data augmentation approach hard sample mining strategy are incompatible ViTs-based FR backbone due lack of tailored consideration on preserving...
Recent advancement in personalized image generation have unveiled the intriguing capability of pre-trained text-to-image models on learning identity information from a collection portrait images. However, existing solutions are vulnerable producing truthful details, and usually suffer several defects such as (i) The generated face exhibit its own unique characteristics, \ie facial shape feature positioning may not resemble key characteristics input, (ii) synthesized contain warped, blurred...
Adaptive filters based on the maximum correntropy criterion (MCC) exhibit good robustness against impulsive interferences. The commonly used kernel function employed in MCC is Gaussian although it may not be best option all cases. To this end, a generalization of kernel, called q-Gaussian combined with developed using gradient, however leading slow convergence rate. overcome defect rate, paper presents novel conjugate gradient (QMCC-CG) algorithm. proposed QMCC-CG algorithm accelerates and...
RoIPool/RoIAlign is an indispensable process for the typical two-stage object detection algorithm, it used to rescale proposal cropped from feature pyramid generate a fixed size map. However, these maps of local receptive fields will heavily lose global context information. To tackle this problem, we propose novel end-to-end trainable framework, called Global Context Aware (GCA) RCNN, aiming at assisting neural network in strengthening spatial correlation between background and foreground by...
The input ground motion for the seismic design of nuclear power plant (NPP) equipment is required to envelop both response spectrum (RRS) and target spectral density (PSD). However, no unified algorithm exists generate PSD, whether at home or abroad. Based on generation methodology PSD suggested in Section 3.7.1 Standard Review Plan (SRP) U.S. Nuclear Regulatory Commission, this paper proposes an improved method synthesis with original iteration process optimized. proposed applied...