Simultaneous matrix diagonalization is a key subroutine in many machine learning problems, including blind source separation and parameter estimation in latent variable models. Here, we extend joint diagonalization algorithms to low-rank and asymmetric matrices and also provide extensions to the perturbation analysis of these methods. Our results allow joint diagonalization to scale to larger p...

Modi cations to Recursive Auto-Associative Memory are presented, which allow it to store deeper and more complex data structures than previously reported. These modi cations include adding extra layers to the compressor and reconstructor networks, employing integer rather than real-valued representations, pre-conditioning the weights and pre-setting the representations to be compatible with the...

Recent years have witnessed a growing interest in analogical learning for NLP applications. If the principle of analogical learning is quite simple, it does involve complex steps that seriously limit its applicability, the most computationally demanding one being the identification of analogies in the input space. In this study, we investigate different strategies for efficiently solving this p...

Some educational interventions successfully “scale up.” Others do not. Little— arguably, almost nothing—is known about the factors that lead to successful scaling up. The goal of this chapter is to identify a number of these factors through a disciplined and methodologically rigorous approach. The difficulties associated with scaling up can broadly be summarized as falling into two classes: (1)...

Establishing the relative value of results within a field of study contributes to advancement of that field. In order to compare across large numbers of results together, the methods and metrics used must be scaled up from existing studies where the number of subjects or cases is small. This scaling provides specific challenges, for example, metrics used in small studies are often shaped by fac...

This paper briefly summarizes definitions, approaches, and challenges to achieving “scale” in community-focused health programs as discussed at the 2005 CORE spring meeting and the USAID child survival and health grants program mini-university. This paper is meant to harmonize a vocabulary for use by NGOs and their partners as they further discuss, debate, and analyze how NGOs and their partner...

As standardly implemented in R or the Tetrad program, causal search algorithms used most widely or effectively by scientists have severe dimensionality constraints that make them inappropriate for big data problems without sacrificing accuracy. However, implementation improvements are possible. We explore optimizations for the Greedy Equivalence Search that allow search on 50,000-variable probl...

Word clusters improve performance in many NLP tasks including training neural network language models, but current increases in datasets are outpacing the ability of word clusterers to handle them. In this paper we present a novel bidirectional, interpolated, refining, and alternating (BIRA) predictive exchange algorithm and introduce ClusterCat, a clusterer based on this algorithm. We show tha...

Scaling up video resolution has conventionally been achieved via linear interpolation, however this method occasionally introduces blurring to the output. Superresolution (SR), an approach to preserve image quality in enlarged still images, has been exploited as a substitute for linear interpolation, however, the output at times exhibits image qualities worse than what linear interpolation prod...