Amazon.com

About: Amazon.com is a company organization based out in Seattle, Washington, United States. It is known for research contribution in the topics: Computer science & Service (business). The organization has 13363 authors who have published 17317 publications receiving 266589 citations.

A survey of faceted search

Abstract: Faceted Search is an exploratory search mechanism, which provides an iterative way to refine search results by a faceted taxonomy. With the benefit of search results diversification, no need for a priori knowledge, and never leading to zero result, it can significantly reduce information overload. Faceted Search has witnessed a booming interest in the last ten years. In this paper, we first analyze the representative facet search models. Next, we present a general faceted search framework, and survey the related methods and techniques, including facet term extraction, hierarchy construction, compound term generation and facet ranking. Then we discuss the metrics for faceted search evaluation, and also highlight the main characteristics of a number of existing faceted search systems. Some directions for future research are finally presented.

Method and system for providing a discount at an auction

Abstract: A method and system for conducting an auction. The auction system provides a mechanism that allows the winning bidder to receive a discount from the winning bid amount when a certain discount criteria is met. The discount criteria is met when the winning bidder was the first bidder to place the bid at the auction. That is, the bidder who places the first bid will receive a discount (e.g., 10%) if that bidder is ultimately successful in winning the auction. The mechanism, tracks whether each auction is subject to a “first bidder discount” and whether the winning bidder was the first bidder. The offering of a first bidder discount and the amount of any discount may be at the discretion of the seller of the item.

User input with fingerprint sensor

Abstract: Devices such as tablets, smartphones, media players, and so forth may incorporate a fingerprint sensor to support acquisition of biometric identification data. As described herein, input data from the fingerprint sensor may be used to control one or more functions of the device. The function controlled may be based at least in part on context of one or more applications executing on the device, direction of motion, and so forth. In one implementation, movement parallel to the fingerprint sensor may modify audio volume settings on the device.

signSGD with Majority Vote is Communication Efficient and Fault Tolerant

Abstract: Training neural networks on large datasets can be accelerated by distributing the workload over a network of machines. As datasets grow ever larger, networks of hundreds or thousands of machines become economically viable. The time cost of communicating gradients limits the effectiveness of using such large machine counts, as may the increased chance of network faults. We explore a particularly simple algorithm for robust, communication-efficient learning---signSGD. Workers transmit only the sign of their gradient vector to a server, and the overall update is decided by a majority vote. This algorithm uses 32× less communication per iteration than full-precision, distributed SGD. Under natural conditions verified by experiment, we prove that signSGD converges in the large and mini-batch settings, establishing convergence for a parameter regime of Adam as a byproduct. Aggregating sign gradients by majority vote means that no individual worker has too much power. We prove that unlike SGD, majority vote is robust when up to 50% of workers behave adversarially. The class of adversaries we consider includes as special cases those that invert or randomise their gradient estimate. On the practical side, we built our distributed training system in Pytorch. Benchmarking against the state of the art collective communications library (NCCL), our framework---with the parameter server housed entirely on one machine---led to a 25% reduction in time for training resnet50 on Imagenet when using 15 AWS p3.2xlarge machines.

Storm intensity and old-growth forest disturbances in the Amazon region

Abstract: [1] We analyzed the pattern of large forest disturbances or blow-downs apparently caused by severe storms in a mostly unmanaged portion of the Brazilian Amazon using 27 Landsat images and daily precipitation estimates from NOAA satellite data. For each Landsat a spectral mixture analysis (SMA) was applied. Based on SMA, we detected and mapped 279 patches (from 5 ha to 2,223 ha) characteristic of blow-downs. A total of 21,931 ha of forest were disturbed. We found a strong correlation between occurrence of blow-downs and frequency of heavy rainfall (Spearman's rank, r2 = 0.84, p < 0.0003). The recurrence intervals of large disturbances were estimated to be 90,000 yr for the eastern Amazon and 27,000 yr for the western Amazon. This suggests that weather patterns affect the frequency of large forest disturbances that may produce different rates of forest turnover in the eastern and western Amazon basin.