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: Service (business) & Service provider. The organization has 13363 authors who have published 17317 publications receiving 266589 citations.

Deep Gaussian processes for multi-fidelity modeling

Abstract: Multi-fidelity methods are prominently used when cheaply-obtained, but possibly biased and noisy, observations must be effectively combined with limited or expensive true data in order to construct reliable models. This arises in both fundamental machine learning procedures such as Bayesian optimization, as well as more practical science and engineering applications. In this paper we develop a novel multi-fidelity model which treats layers of a deep Gaussian process as fidelity levels, and uses a variational inference scheme to propagate uncertainty across them. This allows for capturing nonlinear correlations between fidelities with lower risk of overfitting than existing methods exploiting compositional structure, which are conversely burdened by structural assumptions and constraints. We show that the proposed approach makes substantial improvements in quantifying and propagating uncertainty in multi-fidelity set-ups, which in turn improves their effectiveness in decision making pipelines.

Understanding Mixup Training Methods

Abstract: Mixup is a neural network training method that generates new samples by linear interpolation of multiple samples and their labels. The mixup training method has better generalization ability than the traditional empirical risk minimization method (ERM). But there is a lack of a more intuitive understanding of why mixup will perform better. In this paper, several different sample mixing methods are used to test how neural networks learn and infer from mixed samples to illustrate how mixups work as a data augmentation method and how it regularizes neural networks. Then, a method of weighting noise perturbation was designed to visualize the loss functions of mixup and ERM training methods to analyze the properties of their high-dimensional decision surfaces. Finally, by analyzing the mixture of samples and their labels, a spatial mixup approach was proposed that achieved the state-of-the-art performance on the CIFAR and ImageNet data sets. This method also enables the generative adversarial nets to have more stable training process and more diverse sample generation ability.

Autonomous delivery transportation network

Abstract: Aspects of an autonomous delivery transportation network for the delivery of items are described. The network includes a central management system that directs the operation of a network of vehicles for delivering items. In one embodiment, the system receives a request for transportation including a pickup location, a delivery location, and item attribute data. The system analyzes the existing service routes of vehicles in the network to identify a vehicle compatible with the request. The system also estimates a delay to the existing service route of the vehicle with reference to the pickup and delivery locations and, when determining that the delay is acceptable, assigns the vehicle to service the request. When determining that the delay is unacceptable, the system may dispatch a new vehicle to join the network. The system may also communicate various aspects of service, such as estimated pickup and/or drop off times, to client devices.

DeCoAR 2.0: Deep Contextualized Acoustic Representations with Vector Quantization

Abstract: Recent success in speech representation learning enables a new way to leverage unlabeled data to train speech recognition model. In speech representation learning, a large amount of unlabeled data is used in a self-supervised manner to learn a feature representation. Then a smaller amount of labeled data is used to train a downstream ASR system using the new feature representations. Based on our previous work DeCoAR and inspirations from other speech representation learning, we propose DeCoAR 2.0, a Deep Contextualized Acoustic Representation with vector quantization. We introduce several modifications over the DeCoAR: first, we use Transformers in encoding module instead of LSTMs; second, we introduce a vector quantization layer between encoder and reconstruction modules; third, we propose an objective that combines the reconstructive loss with vector quantization diversity loss to train speech representations. Our experiments show consistent improvements over other speech representations in different data-sparse scenarios. Without fine-tuning, a light-weight ASR model trained on 10 hours of LibriSpeech labeled data with DeCoAR 2.0 features outperforms the model trained on the full 960-hour dataset with filterbank features.

Fast, Automated, Scalable Generation of Textured 3D Models of Indoor Environments

Abstract: 3D modeling of building architecture from mobile scanning is a rapidly advancing field. These models are used in virtual reality, gaming, navigation, and simulation applications. State-of-the-art scanning produces accurate point-clouds of building interiors containing hundreds of millions of points. This paper presents several scalable surface reconstruction techniques to generate watertight meshes that preserve sharp features in the geometry common to buildings. Our techniques can automatically produce high-resolution meshes that preserve the fine detail of the environment by performing a ray-carving volumetric approach to surface reconstruction. We present methods to automatically generate 2D floor plans of scanned building environments by detecting walls and room separations. These floor plans can be used to generate simplified 3D meshes that remove furniture and other temporary objects. We propose a method to texture-map these models from captured camera imagery to produce photo-realistic models. We apply these techniques to several data sets of building interiors, including multi-story datasets.