Feng Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Feng Zhao has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Donglin Wang
Westlake University, Hangzhou, CHINA
2 shared publications - Tiancheng Huang
College of Computing and Data Science, Nanyang Technological University
2 shared publications - Andrew T. Campbell
Center for Technology and Behavioral Health, Geisel School of Medicine, Dartmouth College
1 shared publication Donglin Wang
School of Engineering, Westlake University, Hangzhou, China
1 shared publication- Hong Lu
Intel Labs Santa 95054 Clara CA USA
1 shared publication - Min Zhang
East China Normal University, Shanghai, China, and MBZUAI, Abu Dhabi, United Arab Emirates
1 shared publication
Latest Publications
- Open Access
- 12 June 2023
- Published by Association for Computing Machinery (ACM)
- p. 462-471
TAGM: Task-Aware Graph Model for Few-shot Node Classification
- 12 June 2023
- Published by Association for Computing Machinery (ACM)
- p. 462-471
Open AccessTAGM: Task-Aware Graph Model for Few-shot Node Classification
- 8 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (1) , 1415-1422
Graph Few-Shot Learning via Restructuring Task Graph
Existing graph few-shot learning (FSL) methods usually train a model on many task graphs and transfer the learned model to a new task graph. However, the task graphs often contain a great number of isolated nodes, which results in the severe deficiency of learned node embeddings. Furthermore, in the training process, the neglect of task information also constrains the model’s expressive ability. In this brief, we propose a novel metric-based graph few-shot learning approach via restructuring task graph (GFL-RTG). To solve the problems above, we innovatively restructure the task graph by adding class nodes and a task node to the original individual task graph. We first add class nodes and determine the connectivity between class nodes and others via their similarity. Then, we utilize a graph pooling network to learn a task embedding, which is regarded as a task node. Finally, the new task graph is restructured by combining class nodes, task node, and original nodes, which is then used as input to the metric-based graph neural network (GNN) to conduct few-shot learning. Our extensive experiments on three graph-structured datasets demonstrate that our proposed method generally outperforms the state-of-the-art baselines in few-shot learning.
- 8 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (1) , 1415-1422
Graph Few-Shot Learning via Restructuring Task Graph
Existing graph few-shot learning (FSL) methods usually train a model on many task graphs and transfer the learned model to a new task graph. However, the task graphs often contain a great number of isolated nodes, which results in the severe deficiency of learned node embeddings. Furthermore, in the training process, the neglect of task information also constrains the model’s expressive ability. In this brief, we propose a novel metric-based graph few-shot learning approach via restructuring task graph (GFL-RTG). To solve the problems above, we innovatively restructure the task graph by adding class nodes and a task node to the original individual task graph. We first add class nodes and determine the connectivity between class nodes and others via their similarity. Then, we utilize a graph pooling network to learn a task embedding, which is regarded as a task node. Finally, the new task graph is restructured by combining class nodes, task node, and original nodes, which is then used as input to the metric-based graph neural network (GNN) to conduct few-shot learning. Our extensive experiments on three graph-structured datasets demonstrate that our proposed method generally outperforms the state-of-the-art baselines in few-shot learning.
- Open Access
- 15 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 116348-116357
A Probabilistic Approach for WiFi Fingerprint Localization in Severely Dynamic Indoor Environments
Taking advantage of widely deployed access points (AP), WiFi fingerprint based localization is of importance in indoor internet-of-things (IOT) environments. Nevertheless, spatio-temporal variation is one of its intractable problems, indicating severely environmental dynamics and uncertainty of decision. In this case, the localization accuracy drops significantly. In this paper, we attempt to overcome effects of spatio-temporal variations from two aspects: filtering of the training data and selection of partially valuable APs for matching in test phase. The key idea is to match partial unaffected measurements with ‘clean’ unaffected fingerprints. Bayesian framework and category model are presented for WiFi fingerprints. Two binary hidden variables with different dimensions are introduced to identify singular fingerprints and affected measurements respectively by employing expectation-maximization (EM) algorithms. EM based filter and simultaneous AP selection and localization are then proposed to obtain an optimal matching. Experimental results show that our proposed scheme greatly improves the localization accuracy in severely dynamic indoor environments.
- 15 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 116348-116357
Open AccessA Probabilistic Approach for WiFi Fingerprint Localization in Severely Dynamic Indoor Environments
Taking advantage of widely deployed access points (AP), WiFi fingerprint based localization is of importance in indoor internet-of-things (IOT) environments. Nevertheless, spatio-temporal variation is one of its intractable problems, indicating severely environmental dynamics and uncertainty of decision. In this case, the localization accuracy drops significantly. In this paper, we attempt to overcome effects of spatio-temporal variations from two aspects: filtering of the training data and selection of partially valuable APs for matching in test phase. The key idea is to match partial unaffected measurements with ‘clean’ unaffected fingerprints. Bayesian framework and category model are presented for WiFi fingerprints. Two binary hidden variables with different dimensions are introduced to identify singular fingerprints and affected measurements respectively by employing expectation-maximization (EM) algorithms. EM based filter and simultaneous AP selection and localization are then proposed to obtain an optimal matching. Experimental results show that our proposed scheme greatly improves the localization accuracy in severely dynamic indoor environments.
- 22 May 2013
- Published by Springer Nature in Personal and Ubiquitous Computing
- Vol. 18 (2) , 355-368
Community Similarity Networks
- 22 May 2013
- Published by Springer Nature in Personal and Ubiquitous Computing
- Vol. 18 (2) , 355-368
Community Similarity Networks