In a deep neural network (DNN), the number of parameters is usually huge to get high learning performances. For that reason, it costs lot memory and substantial computational resources, also causes overfitting. It known some are redundant can be removed from without decreasing performance. Many sparse regularization criteria have been proposed solve this problem. convolutional (CNN), group regularizations often used remove unnecessary subsets weights, such as filters or channels. When we...

A hierarchical floorplanning method is presented in which it possible to determine simultaneously a floorplan and detailed global routes directly correspond switchboxes channels. In this method, since the precise estimation of congestion channels possible, no extremely congested are created. Experimental results show that can produce an optimal terms congestion.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

In this work, we will consider the dimension reduction of set time series, such as economic data, to find meaningful basis vector for and indicate which data use vector. Usually each series is analyzed independently in economics but here analyze simultaneously. Since some are measured positive values want decompose them a mixture parts, apply non-negative matrix factorization data. Non-negative can compress dimensions by approximating with product two matrices. The matrices called...

The dimensionality reduction has been widely introduced to use the high-dimensional data for regression, classification, feature analysis, and visualization. As one technique of reduction, a stochastic neighbor embedding (SNE) was introduced. SNE leads powerful results visualize by considering similarity between local Gaussian distributions high low-dimensional space. To improve SNE, t-distributed (t-SNE) also data, t-SNE more flexible visualization on 2 or 3-dimensional mapping than using...

Deep learning has been successfully applied to a variety of tasks. However, lot unnecessary information for the task is included in training data and it difficult automatically remove such trained model. For example, necessary ignore variations patients measured medical diagnosis. To address this problem, we propose method that applies module called Gradient Reversal Layer train model by removing extracting only relevant target task. In study, conducted three experiments demonstrate...

This paper attempts to design a user-friendly interface which enhances usability of relational database systems. Our approach is semantic data model suitable for user and provide an easy-to-use query language DOMCALC (domain calculus language) based on the model. A in can be described terms intuitive notions, namely domains, instead relations attributes. That is, requires users no knowledge conceptual such as relation names attribute names. Therefore, removes requirement from protects views...

The behavior of complex software may be changed depending on the external environment such as user profile, available resources, input data, time, and so on. Design dynamic in conventional languages needs extra effort to keep consistency among behaviors for all states. We propose an object oriented visual language its visualized specification with dynamics presentation. supports according both time-dependent time-independent states based objects. In specification, we present uniform...

A deep convolutional neural network (CNN) has been widely used in image classification and gives better accuracy than the other techniques. The softmax cross-entropy loss function is often for tasks. There are some works to introduce additional terms objective training make features of output layer more discriminative. neuron-wise discriminant criterion makes input feature each neuron discriminative by introducing features. Similarly, center was introduced before activation face recognition...

A visualized and modeless programming environment VPF (Visualized Programming for Form manipulation language) is presented. The supports the development of programs written in a language called form language. so-called iconic provides not only facilities but also form-oriented operations. following features: (1) hierarchical views that support various abstract levels program specification execution; (2) visualization itself its run-time behavior with graphical representation; (3) operations...

The problem of pipelined data path synthesis in VLSI design is investigated. Given the behavioral description circuit to be synthesized, which one or more conditional branches may nested, authors propose a heuristic algorithm, takes mutual exclusiveness into account so that than operation shares hardware module, and hence chip with small size will produced. Experimental results show proposed algorithm efficiently produces nearly optimum solution. present resource sharing simultaneously...

A basic framework for the design of a prototype form system, CRAFT, is proposed. CRAFT enables users to create (i.e. define and realize) forms with various structures. Form creation an integrated concept definition realization. For definition, formal model introduced. realization specified by using graph based on AND/OR trees. supports visual language that provides naive easy programming method.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

The task of image annotation is becoming enormously important for efficient retrieval from the web and other large databases. However, huge semantic information complex dependency labels on an make challenging. Hence determining similarity between multiple useful to understand any incomplete label assignment retrieval. This work proposes a novel method solve problem multi-label by unifying two different types Laplacian regularization terms in deep convolutional neural network (CNN) robust...

The dimensionality reduction has been widely introduced to use the high-dimensional data for regression, classification, feature analysis, and visualization. As one technique of reduction, a stochastic neighbor embedding (SNE) was introduced. SNE leads powerful results visualize by considering similarity between local Gaussian distributions high low-dimensional space. To improve SNE, t-distributed (t-SNE) also data, t-SNE more flexible visualization on 2 or 3-dimensional mapping than using...

Abstract This paper considers the building block method, which is one of methods in VLSI layout design, and discusses improvement module placement assuming that a one‐dimensional modules given as initial placement. Discussions are made following: The case where net list input composed only two‐terminal nets, connect external terminals modules. problem PMO formulated, virtual connection length defined by × coordinate to be minimized while retaining order algorithm solve this problem, well...

Embedded microprocessors are widely used in many kinds of industrial products, and the number processors a product has been greatly increased. Thus, such system constitutes complex embedded which is distributed reliable. I propose software architecture for systems based on concepts autonomous error checking correction, acceptance test with load control, request deletion. They will resolve problems unexpected overload, interference from other processors, no protection failure or bugs.

This paper proposes a two-step detector called segmented object detection, whose performance is improved by masking the background region. Previous single-stage detection methods suffer from problem of imbalance between foreground and classes, where occupies more regions in image than foreground. Thus, loss firmly incorporated into training. RetinaNet addresses this with Focal Loss, which focuses on loss. Therefore, we propose method that generates probability maps using instance...

Abstract In VLSI layout design by the building block approach, a hierarchical floorplanning method is proposed in which floorplan and global routing are determined simultaneously to realize low‐congestion switchboxes. this method, overlap resolution, preserves relative position minimizes obtained chip area, needed prevent increase wire length of each net. paper, fundamental study resolution made, following described. (1) The formulation problem RI given. formulation, (a) preservation among...