Residual networks (Resnets) have become a prominent architecture in deep learning. However, a comprehensive understanding of Resnets is still a topic of ongoing research. A recent view argues that Resnets perform iterative refinement of features. We attempt to further expose properties of this aspect. To this end, we study Resnets both analytically and empirically. We formalize the notion of it...

The focus of this article is distillation design feasibility. It is shown that existing methods for determining feasibility can give incorrect results or produce feasible designs that waste energy due to over-specification and mass balance errors. An iterative refinement procedure based on direct substitution is proposed within the distillation line method of Lucia et al. that automatically adj...

In a cell, it has been estimated that each protein on average interacts with roughly 10 others, resulting in tens of thousands of proteins known or suspected to have interaction partners; of these, only a tiny fraction have solved protein structures. To partially address this problem, we have developed M-TASSER, a hierarchical method to predict protein quaternary structure from sequence that in...

Iterative localization is designed to more free nodes when the number of anchor is few. When all localizable nodes are localized in the primitive iterative localization, the reciprocal refinement localization is proposed to refine and improve the node positions. To improve the localization accuracy, the position error of pseudo anchor is transformed to the equivalent range error, the optimal we...

A planning problem is time-dependent, if the time spent planning affects the uti l i ty of the system's performance. In [Dean and Boddy, 1988], we define a framework for constructing solutions to time-dependent planning problems, called expectation-driven iterative refinement. In this paper, we analyze and solve a moderately complex time-dependent planning problem involving path planning for a ...

When designing and deploying distributed systems it is necessary to determine process activation policy. A process’s activation policy determines whether it is persistent or should be created and terminated with each call. For persistent processes the replication or threading levels must be decided. Inappropriate replication/threading levels can lead to unnecessary queuing delays for callers or...

We describe a novel approach to comparative assembly that directly integrates anchoring alignments into the contig assembly process, enabling the extension of contig construction through the boundaries of repeat nodes in a compressed de Bruijn graph. Our method exploits anchoring alignments, paired-read constraints and read threading as path selection heuristics while an assembly graph is trans...

We propose an extension of the abstract rules for independent implementability of reactive components proposed in interface theories to take into account interface assemblies. As a concrete instantiation we extend existing interface theories for modal I/O-transition systems to support assemblies, (greybox) assembly refinement and assembly encapsulation. We introduce a new notion of communicatio...