We develop a novel multiple instance SVM based on maximizing the minimum witness rate (WR) among positive bags. We solve this nonlinear integer programming problem by the multiple instance SVM formulation with a convex loss function, where unknown integer labels of instances in positive bags are relaxed to real numbers between -1 and 1 using the tanh(·) to estimate WR of a positive bag. Our new model, WR-SVM, also can incorporates the multiple instance learning (MIL) problem into a simple deep neural network framework with no additional MIL pooling layers. WR-SVM is robust to input perturbation by eliminating the imbalance between positive instances allocated to positive bags. A better generalization ability is expected by the large margin due to the balanced allocation of positive instances to positive bags. We perform experiments on various video datasets to verify the effectiveness of our method for video classification. The results of WR-SVM outperform the state-of-the-art for MIL-based video classification models.

We propose a static routing model for finding an optimal logical network configuration of a given ATM-based B-ISDN. The model is essentially identical to the model proposed by Kim (1995) which finds a virtual path configuration in which an optimal trade-off is achieved among the relevant gains of ATM technology. Unlike Kim's model, however our model explicitly considers the VP's quality of service to enhance the practicality and the bandwidth utilization. We propose a heuristic algorithm tailored to the computational structure of the model. A computational experiment shows that the algorithm finds good solutions in dramatically reduced computation time compared to the existing algorithm. This suggests the possibility of a real-time VP management system in which optimal routing is implemented on-line with dynamic routing, to maintain the optimality of the current solutions.

A number of multicast tree algorithms have been proposed in the literature. Their approach to modeling the problem has some drawbacks in applications for ATM networks since only a single multipoint connection request is considered in the presence of link capacity constraints. In this paper, we present a detailed mathematical model for the generalized multicast arborescence (i.e., a roofed out-tree) problem in which there are a number of concurrent multipoint connections of various traffic types. Point-to-point connections are also considered as a special case of multipoint connections. A solution to the model determines, in the most economical way, an establishment of VPCs with reserved capacity and multicast arborescences using the established VPCs for multipoint connections. This model is a natural extension of the virtual channel assignment problem introduced by Kim (1995). Algorithms and numerical results are presented.

A number of optimization models for the establishment of virtual path connections (VPCs) with reserved capacity have been proposed in the literature. All of them, however, failed to reflect the trade-off between increased capacity costs and reduced control costs explicitly in their formulations. the present authors develop a mathematical formulation for an optimal strategy for the reservation of capacity on VPCs, considering the trade-off between increased capacity costs and reduced control costs. In this formulation, they also consider the reservation of buffers on VPCs. Algorithms and numerical results are presented.