Graduated optimization is a global optimization technique that is used to minimize a multimodal nonconvex function by smoothing the objective function with noise and gradually refining the solution. This paper experimentally evaluates the performance of the explicit graduated optimization algorithm with an optimal noise scheduling derived from a previous study and discusses its limitations. The evaluation uses traditional benchmark functions and empirical loss functions for modern neural network architectures. In addition, this paper extends the implicit graduated optimization algorithm, which is based on the fact that stochastic noise in the optimization process of SGD implicitly smooths the objective function, to SGD with momentum, analyzes its convergence, and demonstrates its effectiveness through experiments on image classification tasks with ResNet architectures.

A scaled conjugate gradient method that accelerates existing adaptive methods utilizing stochastic gradients is proposed for solving nonconvex optimization problems with deep neural networks. It is shown theoretically that, whether with a constant or diminishing learning rate, the proposed method can obtain a stationary point of the problem. Additionally, its rate of convergence with a diminishing learning rate is verified to be superior to that of the conjugate gradient method. The proposed method is shown to minimize training loss functions faster than the existing adaptive methods in practical applications of image and text classification. Furthermore, in the training of generative adversarial networks, one version of the proposed method achieved the lowest Fréchet inception distance score among those of the adaptive methods.

Previous results have shown that a two time-scale update rule (TTUR) using different learning rates, such as different constant rates or different decaying rates, is useful for training generative adversarial networks (GANs) in theory and in practice. Moreover, not only the learning rate but also the batch size is important for training GANs with TTURs and they both affect the number of steps needed for training. This paper studies the relationship between batch size and the number of steps needed for training GANs with TTURs based on constant learning rates. We theoretically show that, for a TTUR with constant learning rates, the number of steps needed to find stationary points of the loss functions of both the discriminator and generator decreases as the batch size increases and that there exists a critical batch size minimizing the stochastic first-order oracle (SFO) complexity. Then, we use the Fréchet inception distance (FID) as the performance measure for training and provide numerical results indicating that the number of steps needed to achieve a low FID score decreases as the batch size increases and that the SFO complexity increases once the batch size exceeds the measured critical batch size. Moreover, we show that measured critical batch sizes are close to the sizes estimated from our theoretical results.

MOTIVATION: Trans-kingdom protein clustering remained difficult because of large sequence divergence between eukaryotes and prokaryotes and the presence of a transit sequence in organellar proteins. A large-scale protein clustering including such divergent organisms needs a heuristic to efficiently select similar proteins by setting a proper threshold for homologs of each protein. Here a method is described using two similarity measures and organism count. RESULTS: The Gclust software constructs minimal homolog groups using all-against-all BLASTP results by single-linkage clustering. Major points include (i) estimation of domain structure of proteins; (ii) exclusion of multi-domain proteins; (iii) explicit consideration of transit peptides; and (iv) heuristic estimation of a similarity threshold for homologs of each protein by entropy-optimized organism count method. The resultant clusters were evaluated in the light of power law. The software was used to construct protein clusters for up to 95 organisms. AVAILABILITY: Software and data are available at http://gclust.c.u-tokyo.ac.jp/Gclust_Download.html.

This paper is concerned with coding theorems for a generalized wiretap channel. In the problem of the wiretap channel it is important to characterize the secrecy capacity, i.e., the maximum achievable rate of information securely transmitted from a sender to a legitimate receiver in the presence of a wiretapper. In our model, channels are not restricted to memoryless channels. We first define the secrecy capacity Csand evaluate Csfrom information-spectrum approach. We give a new upper bound of Csnot including auxiliary random variables and explore conditions under which the obtained upper bound becomes tight for the cascaded wiretap channel

SMIL is a markup language which enables us to describe multimedia contents. This paper proposes a design model of SMIL browser functionality for mobile terminals whose resources are limited. We introduce SMIL Component, which is based on attachable software architecture to a pre-installed generic web browser. This leads to reduce memory size that SMIL Component consumes and bring high adaptability of SMIL Component for various web browsers. We implement SMIL Component and evaluate RAM sizes and presentation delays. As a result, we found out that SMIL Component can be useful for MMS presentations on mobile terminals.

Abstract Summary: A Tcl/Tk-based application called GenoMap is described, a viewer for genome-wide map of microarray expression data within a circular bacterial genome. An interactive interface facilitates easy identification of the expressed region. This software is also used for drawing genome-wide quantitative data. Availability: The program is freely available to academic users as a Tcl/Tk source code at http://www.molbiol.saitama-u.ac.jp/~naoki/GenoMap/ Contact: [email protected]

UNLABELLED: A multiple sequence file converter for common platforms, SISEQ,is described, which performs extraction of DNA sequences that correspond to CDS or RNA field of a large database file as well as subsequent multi-sequence conversions for phylogenetic or molecular biological analysis. Command-line interface as well as a GUI and a script-driven operation mode are provided. AVAILABILITY: The program is freely available to academic users in the form of Macintosh FAT binary, DOS executable, or UNIX source code at http://www.molbiol.saitama-u.ac.jp/ñaoki/ Software.html. CONTACT: [email protected]