Wei Zhuo

Wei Zhuo

Auto-generated profile, based on publicly available information

Wei Zhuo

School of Intelligent Systems Engineering, Shenzhen Campus of Sun Yat-sen University, Shenzhen, Guangdong, China

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
15 May 2024 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
ZZZhaohuan Zhan
Zhaohuan Zhan
JQJinghui Qin
Jinghui Qin
WZWei Zhuo
Wei Zhuo
GTGuang Tan
Guang Tan
A challenging task in embodied artificial intelligence is enabling the robot to carry out a navigational task following natural language instruction. In the task, the navigator needs to understand objects, directions, as well as room types, which serve as landmarks for navigation. Although it is easy to encode objects and directions with an external encoder like an object detector, current navigators struggle to encode room type information properly due to the low accuracy offered by existing classifiers. This inadequacy poses confusion that navigators find difficult to overcome. Even humans may sometimes fail to determine the exact type of a room since multiple room types may exist in one panorama. To mitigate this problem, we propose to encode room type information in a prompt manner. Specifically, we first establish multi-modal, learnable prompt pools containing knowledge of room types. By querying the prompt pools, the navigator can obtain room-type prompts of the current view, and incorporate them into the navigator using a prompt-based learning method. Experimental results on the REVERIE, R2R and SOON datasets demonstrate the effectiveness of our approach.

21 August 2019 in Computational Intelligence and Neuroscience
Verified
Journal Article
Open access
PDF file
WZWei Zhuo
Wei Zhuo
QZQianyi Zhan
Qianyi Zhan
Yuan LiuYuan Liu
Yuan Liu
ZXZhenping Xie
Zhenping Xie
JLJing Lu
Jing Lu
Network embedding (NE), which maps nodes into a low-dimensional latent Euclidean space to represent effective features of each node in the network, has obtained considerable attention in recent years. Many popular NE methods, such as DeepWalk, Node2vec, and LINE, are capable of handling homogeneous networks. However, nodes are always fully accompanied by heterogeneous information (e.g., text descriptions, node properties, and hashtags) in the real-world network, which remains a great challenge to jointly project the topological structure and different types of information into the fixed-dimensional embedding space due to heterogeneity. Besides, in the unweighted network, how to quantify the strength of edges (tightness of connections between nodes) accurately is also a difficulty faced by existing methods. To bridge the gap, in this paper, we propose CAHNE (context attention heterogeneous network embedding), a novel network embedding method, to accurately determine the learning result. Specifically, we propose the concept of node importance to measure the strength of edges, which can better preserve the context relations of a node in unweighted networks. Moreover, text information is a widely ubiquitous feature in real-world networks, e.g., online social networks and citation networks. On account of the sophisticated interactions between the network structure and text features of nodes, CAHNE learns context embeddings for nodes by introducing the context node sequence, and the attention mechanism is also integrated into our model to better reflect the impact of context nodes on the current node. To corroborate the efficacy of CAHNE, we apply our method and various baseline methods on several real-world datasets. The experimental results show that CAHNE achieves higher quality compared to a number of state-of-the-art network embedding methods on the tasks of network reconstruction, link prediction, node classification, and visualization.

5 June 2023 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
WZWei Zhuo
Wei Zhuo
GTGuang Tan
Guang Tan
Graph neural networks (GNNs) have been successful in a variety of graph-based applications. Recently, it is shown that capturing long-range relationships between nodes helps improve the performance of GNNs. The phenomenon is mostly confirmed in a supervised learning setting. In this article, inspired by contrastive learning (CL), we propose an unsupervised learning pipeline, in which different types of long-range similarity information are injected into the GNN model in an efficient way. We reconstruct the original graph in feature and topology spaces to generate three augmented views. During training, our model alternately picks an augmented view, and maximizes an agreement between the representations of the view and the original graph. Importantly, we identify the issue of diminishing utility of the augmented views as the model gradually learns useful information from the views. Hence, we propose a view update scheme that adaptively adjusts the augmented views, so that the views can continue to provide new information that helps with CL. The updated augmented views and the original graph are jointly used to train a shared GNN encoder by optimizing an efficient channel-level contrastive objective. We conduct extensive experiments on six assortative graphs and three disassortative graphs, which demonstrate the effectiveness of our method.

27 September 2021 in arXiv
Verified
Preprint
Open access
PDF file
LYLihe Yang
Lihe Yang
WZWei Zhuo
Wei Zhuo
LQLei Qi
Lei Qi
YSYinghuan Shi
Yinghuan Shi
YGYang Gao
Yang Gao
Few-shot segmentation (FSS) aims to segment unseen classes given only a few annotated samples. Existing methods suffer the problem of feature undermining, i.e. potential novel classes are treated as background during training phase. Our method aims to alleviate this problem and enhance the feature embedding on latent novel classes. In our work, we propose a novel joint-training framework. Based on conventional episodic training on support-query pairs, we add an additional mining branch that exploits latent novel classes via transferable sub-clusters, and a new rectification technique on both background and foreground categories to enforce more stable prototypes. Over and above that, our transferable sub-cluster has the ability to leverage extra unlabeled data for further feature enhancement. Extensive experiments on two FSS benchmarks demonstrate that our method outperforms previous state-of-the-art by a large margin of 3.7% mIOU on PASCAL-5i and 7.0% mIOU on COCO-20i at the cost of 74% fewer parameters and 2.5x faster inference speed. The source code is available at https://github.com/LiheYoung/MiningFSS.