Multi-Graph Multi-Label Learning Based on Entropy

Zhu, Zhaoqi; Zhao, Yuhai

doi:10.3390/e20040245

Cited by 7 publications

(4 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the bag labeling, the following methods are compared: MIMLfast, M3MIML (Zhang and Zhou 2008), HLK, MGMLent (Zhu and Zhao 2018), SGSL SVM and DUMMY. Note that MIMLfast and M3MIML use the bag-of-instances representation for data instead of bag-of-graphs representation.…”

Section: Experiments Baseline Algorithmsmentioning

confidence: 99%

“…For the same reason multi-graph multi-label learning aims to extend MIML to the graph field. Zhu et al (Zhu and Zhao 2018) proposed a MIML algorithm based on entropy to solve MGML through mining the informative subgraph using entropy.…”

Section: Related Workmentioning

confidence: 99%

“…Later on, Multi-graph Multi-label (MGML) learning is proposed to release the constraint of single label issue and to enable the calssification model to describe each bag of graphs with multiple labels. The second class of MGL learning is proven to be more efficient than the MIL and the first class methods (Zhu and Zhao 2018).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Towards Coarse and Fine-grained Multi-Graph Multi-Label Learning

Wang¹,

Zhao²,

Wang³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Multi-graph multi-label learning (MGML) is a supervised learning framework, which aims to learn a multi-label classifier from a set of labeled bags each containing a number of graphs. Prior techniques on the MGML are developed based on transfering graphs into instances and focus on learning the unseen labels only at the bag level. In this paper, we propose a coarse and fine-grained Multi-graph Multi-label (cfMGML) learning framework which directly builds the learning model over the graphs and empowers the label prediction at both the coarse (aka. bag) level and fine-grained (aka. graph in each bag) level. In particular, given a set of labeled multi-graph bags, we design the scoring functions at both graph and bag levels to model the relevance between the label and data using specific graph kernels. Meanwhile, we propose a thresholding rank-loss objective function to rank the labels for the graphs and bags and minimize the hamming-loss simultaneously at one-step, which aims to addresses the error accumulation issue in traditional rank-loss algorithms. To tackle the nonconvex optimization problem, we further develop an effective sub-gradient descent algorithm to handle high-dimensional space computation required in cfMGML. Experiments over various real-world datasets demonstrate cfMGML achieves superior performance than the state-of-arts algorithms.

show abstract

Section: Experiments Baseline Algorithmsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Towards Coarse and Fine-grained Multi-Graph Multi-Label Learning

Wang¹,

Zhao²,

Wang³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…As an important carrier of information, it is significant to do efficient research with images [ 1 , 2 , 3 , 4 , 5 , 6 ]. Large-scale image retrieval has vast applications in many domains such as image analysis, search of image over internet, medical image retrieval, remote sensing, and video surveillance [ 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 ].…”

Section: Introductionmentioning

confidence: 99%

An Adaptive Weight Method for Image Retrieval Based Multi-Feature Fusion

Wang

et al. 2018

Entropy

View full text Add to dashboard Cite

With the rapid development of information storage technology and the spread of the Internet, large capacity image databases that contain different contents in the images are generated. It becomes imperative to establish an automatic and efficient image retrieval system. This paper proposes a novel adaptive weighting method based on entropy theory and relevance feedback. Firstly, we obtain single feature trust by relevance feedback (supervised) or entropy (unsupervised). Then, we construct a transfer matrix based on trust. Finally, based on the transfer matrix, we get the weight of single feature through several iterations. It has three outstanding advantages: (1) The retrieval system combines the performance of multiple features and has better retrieval accuracy and generalization ability than single feature retrieval system; (2) In each query, the weight of a single feature is updated dynamically with the query image, which makes the retrieval system make full use of the performance of several single features; (3) The method can be applied in two cases: supervised and unsupervised. The experimental results show that our method significantly outperforms the previous approaches. The top 20 retrieval accuracy is 97.09%, 92.85%, and 94.42% on the dataset of Wang, UC Merced Land Use, and RSSCN7, respectively. The Mean Average Precision is 88.45% on the dataset of Holidays.

show abstract