Lehigh University

About: Lehigh University is a education organization based out in Bethlehem, Pennsylvania, United States. It is known for research contribution in the topics: Catalysis & Fracture mechanics. The organization has 12684 authors who have published 26550 publications receiving 770061 citations.

Mini-Batch Semi-Stochastic Gradient Descent in the Proximal Setting

Abstract: We propose mS2GD: a method incorporating a mini-batching scheme for improving the theoretical complexity and practical performance of semi-stochastic gradient descent (S2GD). We consider the problem of minimizing a strongly convex function represented as the sum of an average of a large number of smooth convex functions, and a simple nonsmooth convex regularizer. Our method first performs a deterministic step (computation of the gradient of the objective function at the starting point), followed by a large number of stochastic steps. The process is repeated a few times with the last iterate becoming the new starting point. The novelty of our method is in introduction of mini-batching into the computation of stochastic steps. In each step, instead of choosing a single function, we sample $b$ functions, compute their gradients, and compute the direction based on this. We analyze the complexity of the method and show that it benefits from two speedup effects. First, we prove that as long as $b$ is below a certain threshold, we can reach any predefined accuracy with less overall work than without mini-batching. Second, our mini-batching scheme admits a simple parallel implementation, and hence is suitable for further acceleration by parallelization.

Greater drought intensity and frequency before AD 1200 in the Northern Great Plains, USA

Abstract: EXTREME large-scale droughts in North America, such as the 'Dust Bowl' of the 1930s, have been infrequent events within the documented history of the past few hundred years, yet this record may not be representative of long-term patterns of natural variation of drought intensity and frequency. In the Great Plains region of central North America, historical droughts have persisted longer than in any other part of the United States1, but no detailed records of drought patterns in this region have hitherto been obtained that extend beyond the past 500 years. Here we present a reconstruction of drought intensity and frequency over the past 2,300 years in the Northern Great Plains, based on lake salinity fluctuations inferred from fossil diatom assemblages. This record, of sub-decadal resolution, suggests that extreme droughts of greater intensity than that of the 1930s were more frequent before AD 1200. This high frequency of extreme droughts persisted for centuries, and was most pronounced during AD 200–370, AD 700–850 and AD 1000–1200. We suggest that before AD 1200, the atmospheric circulation anomalies that produce drought today were more frequent and persistent.

When the former ceo stays on as board chair: effects on successor discretion, strategic change, and performance

Abstract: Prior research on CEO succession has omitted consideration of a critical institutional reality: some exiting CEOs do not fully depart the scene but instead remain as board chairs. We posit that predecessor retention restricts a successor's discretion, thus dampening his or her ability to make strategic changes or deliver performance that deviates from pre-succession levels. In short, a predecessor's continuing presence suppresses a new CEO's influence. Based on analysis of 181 successions in high technology firms, and with extensive controls (for circumstances associated with succession, the firm's need and capacity for change, and for endogeneity), we find substantial support for our hypotheses. In supplementary analyses, we find that retention has a more pronounced effect in preventing a new CEO from making big performance gains than in preventing big drops. Copyright © 2011 John Wiley & Sons, Ltd.

Communication-Efficient Distributed Dual Coordinate Ascent

Abstract: Communication remains the most significant bottleneck in the performance of distributed optimization algorithms for large-scale machine learning. In this paper, we propose a communication-efficient framework, COCOA, that uses local computation in a primal-dual setting to dramatically reduce the amount of necessary communication. We provide a strong convergence rate analysis for this class of algorithms, as well as experiments on real-world distributed datasets with implementations in Spark. In our experiments, we find that as compared to state-of-the-art mini-batch versions of SGD and SDCA algorithms, COCOA converges to the same .001-accurate solution quality on average 25 × as quickly.

Crowded minds: The implicit bystander effect.

Abstract: Five studies merged the priming methodology with the bystander apathy literature and demonstrate how merely priming a social context at Time 1 leads to less helping behavior on a subsequent, completely unrelated task at Time 2. In Study 1, participants who imagined being with a group at Time 1 pledged significantly fewer dollars on a charity-giving measure at Time 2 than did those who imagined being alone with one other person. Studies 2-5 build converging evidence with hypothetical and real helping behavior measures and demonstrate that participants who imagine the presence of others show facilitation to words associated with unaccountable on a lexical decision task. Implications for social group research and the priming methodology are discussed.