TAILIEUCHUNG - Using sum match kernel with balanced label tree for large scale image classification

In addition, a feature map is used to reformulate the sum-match kernel function as a dot product of two mean feature vectors in a mapped-feature space. Furthermore, we proposed an algorithm for learning a balanced tree which gains the computational efficiency in classification. We carried out experiments on benchmark datasets including Caltech-256, SUN-397, and ImageNet-1K. The evaluation results indicated that our method achieves a significant improvement in terms of accuracy and efficiency compared to other methods. In particular, our method achieved in accuracy on ImageNet-1K, compared to of the Bengio et al.’s method. | Journal of Computer Science and Cybernetics, , (2016), 133–152 DOI no. USING SUM MATCH KERNEL WITH BALANCED LABEL TREE FOR LARGE-SCALE IMAGE CLASSIFICATION TIEN-DUNG MAI University of Information Technology, Vietnam National University-Ho Chi Minh City dungmt@ Abstract. Large-scale image classification is a fundamental problem in computer vision due to many real applications in various domains. One of the popular methods is to train one-versus-all binary classifiers independently for each class. Although this method is simple, they are impracticable in the case of a large number of classes because their testing complexity grows linearly with the number of classes. Another is to organize classes into a hierarchical tree structure. The number of classifier evaluations of a test sample when traveling from the root to a leaf node is significantly reduced. A challenging issue is how to learn a tree structure which achieves both classification accuracy and computational efficiency. The current methods use a confusion matrix and spectral clustering techniques to group confusing classes into clusters associated with the nodes. However, training one-vs-all classifiers used to calculate the confusion matrix is costly for a large number of classes. Moreover, the output tree might not be balanced because the objective function of spectral clustering penalizes unbalanced partitions. In this paper, we suggested a novel method to learn the tree structure by using a spectral clustering algorithm and a similarity matrix. Here, the similarity between two classes is exactly measured by the sum-match kernel. In addition, a feature map is used to reformulate the sum-match kernel function as a dot product of two mean feature vectors in a mapped-feature space. Furthermore, we proposed an algorithm for learning a balanced tree which gains the computational efficiency in classification. We carried out experiments on benchmark datasets .

• Kỹ thuật lập trình
• Journal of Computer Science and Cybernetics
• Using sum match kernel
• Balanced label tree for large scale image classification
• Balanced label tree
• The sum match kernel function
• Nonstandard finite difference schemes
• Solving a modified epidemiological model for computer viruses
• Nonstandard finite difference
• Computer virus propagation model
• Multi objective evolutionary algorithms
• Evolutionary algorithms and foundation
• development and open issues
• EC enables computer scientists
• Evolutionary algorithms designed
• Learning interaction measure
• Relevance feedback in image retrieval
• Multi feature query
• The fuzzy density values
• Quality of transmission aware routing in ad hoc networks
• Cross layer model combined
• The static agent
• Ad hoc networks
• QoT aware routing
• sattic agent
• Algorithm to build fuzzy decision tree for data classification problem
• Fuzziness intervals matching
• Fuzzy decision tree
• The fuzzy data mining problem
• Hedges algebras and fuzzy partition problem for qualitative attributes
• Construct the membership functions
• Fuzzy Association Rules
• Member ship functions
• Characterizations for several classes of alternative codes
• Several classes of alternative codes
• norm alternative code
• Left norm alternative code
• Two sided alternative code
• Left alternative code
• Strict alternative code
• Packet classification algorithm
• Multi way priority trie
• Significant number of researchers
• Both theoretically and experimentally proved
• Solving min max capacitated vehicle routing problem by local search
• Solving min max capacitated vehicle routing problem
• Solving min max capacitated
• Proposed algorithms compared
• Novel control approach for optimal power flow
• Hybrid wind photovoltaic diesel generation systems
• Diesel generation systems
• Analytical control model
• Mobile ad hoc network
• Valid route testing mechanism
• Using Network Simulator
• Random movement network topology
• Rough set theory
• Polynomial time complexity
• Algorithm finding object reducts
• Improved genetic algorithm
• Test data generation for simulink models
• Conventional genetic algorithm
• Generate test data for Simulink models
• Nonholonomic wheeled mobile robot
• Wheeled mobile robot
• Function neural network
• Unknown external disturbances
• Modeling the differential motion of a mobile manipulator
• Designing a new visual servoing
• Tracking a flying target
• Reinforcement learning algorithm
• Efficient navigation framework
• Autonomous mobile robots
• The mobile robots
• Interpolation based on semantic distance weighting
• Hedge algebra and its application
• Semantic distance weighting
• Bi linear interpolation
• Linguistic rule base system
• Transform the current Web
• Conceptual data model
• Ti meER model
• The temporal databases into OWL ontology
• The ER model into ontology
• Sliding mode control
• Sliding mode control for a planar parallel robot driven
• Electric motors in a task space
• The substructure method
• Greedy Randomized Adaptive Search Procedure
• Multiple Traveling Repairmen Problem with Distance Constraints
• Variable Neighborhood Descent
• Traveling Repairman Problem
• On the real world semantics interpretability
• Fuzzy rule based systems under fuzzy set approach
• Hedge algebra approach
• The RWS interpretability
• Alternating current electrical systems
• Integer quadratic programming
• Non linear knapsack problems arising
• Polynomial time approximation schemes
• Statistical implication rules
• Collaborative filtering recommender system
• Statistical implicative analysis
• The accuracy of recommender systems
• Picture fuzzy sets
• Picture fuzzy t norms
• Picture fuzzy t conorm
• De Morgan picture operator triple
• Taxonomic assignment for large scale metagenomic data
• High perfomance systems
• Large scale metagenomic data
• Achieve good performance