Forecasts help businesses allocate resources and achieve objectives. At LinkedIn, product owners use forecasts to set business targets, track outlook, and monitor health. Engineers use forecasts to efficiently provision hardware. Developing a forecasting solution to meet these needs requires accurate and interpretable forecasts on diverse time series with sub-hourly to quarterly frequencies. We present Greykite, an open-source Python library for forecasting that has been deployed on over twenty use cases at LinkedIn. Its flagship algorithm, Silverkite, provides interpretable, fast, and highly flexible univariate forecasts that capture effects such as time-varying growth and seasonality, autocorrelation, holidays, and regressors. The library enables self-serve accuracy and trust by facilitating data exploration, model configuration, execution, and interpretation. Our benchmark results show excellent out-of-the-box speed and accuracy on datasets from a variety of domains. Over the past two years, Greykite forecasts have been trusted by Finance, Engineering, and Product teams for resource planning and allocation, target setting and progress tracking, anomaly detection and root cause analysis. We expect Greykite to be useful to forecast practitioners with similar applications who need accurate, interpretable forecasts that capture complex dynamics common to time series related to human activity.

The advent of content-rich modern web applications, unreliable network connectivity and device heterogeneity demands flexible web content delivery platforms that can handle the high variability along many dimensions - especially for the mobile web. Images account for more than 60% of the content delivered by present-day webpages and have a strong influence on the perceived webpage latency and end-user experience. We present a flexible web delivery platform with a client-cloud architecture and content-aware optimizations to address the problem of delivering image-rich web applications. Our solution makes use of quantitative measures of image perceptual quality, machine learning algorithms, partial caching and opportunistic client-side choices to efficiently deliver images on the web. Using data from the WWW, we experimentally demonstrate that our approach shows significant improvement on various web performance criteria that are critical for maintaining a desirable end-user quality-of-experience (QoE) for image-rich web applications.

Due to rapid growth in the richness of web applications and the multitude of wireless devices on which web content can be consumed, optimizing the content delivery of dynamic web applications represents an important technical challenge. One of the keys to solving the wireless web performance puzzle is that of efficient image content delivery. Since the critical factor in image delivery is the end-user experience, we propose a simple quantitative metric for characterizing the Quality of Experience (QoE) for any given image that gets sent through a web delivery service (WDS) pipeline. This quantitative signature, termed VoQS (\emph{variation of quality signature}), allows any two arbitrary images to be compared in the context of web delivery performance. We then use VoQS in conjunction with an unsupervised learning algorithm to group similarly performing images into coherent groups for optimized web delivery. Using a large database of images compiled from multiple content providers and diverse device-optimized formats, we demonstrate that our approach allows large image databases to be efficiently parsed into coherent groups in a content-dependent and device-targeted manner for optimized image content delivery. Our approach significantly decreases the average bits per image that need to be delivered across large image databases (by 43\% in our experiments) while preserving the perceptual quality across the entire image database. We also discuss how such a categorization approach can be leveraged for real-time web delivery of novel image data.

BACKGROUND: Segmenting electron microscopy (EM) images of cellular and subcellular processes in the nervous system is a key step in many bioimaging pipelines involving classification and labeling of ultrastructures. However, fully automated techniques to segment images are often susceptible to noise and heterogeneity in EM images (e.g. different histological preparations, different organisms, different brain regions, etc.). Supervised techniques to address this problem are often helpful but require large sets of training data, which are often difficult to obtain in practice, especially across many conditions. RESULTS: We propose a new, principled unsupervised algorithm to segment EM images using a two-step approach: edge detection via salient watersheds following by robust region merging. We performed experiments to gather EM neuroimages of two organisms (mouse and fruit fly) using different histological preparations and generated manually curated ground-truth segmentations. We compared our algorithm against several state-of-the-art unsupervised segmentation algorithms and found superior performance using two standard measures of under-and over-segmentation error. CONCLUSIONS: Our algorithm is general and may be applicable to other large-scale segmentation problems for bioimages.

Smart camera networks have recently emerged as a new class of sensor network infrastructure that is capable of supporting high-power in-network signal processing and enabling a wide range of applications. In this article, we provide an exposition of our efforts to build a low-bandwidth wireless camera network platform, called CITRIC, and its applications in smart camera networks. The platform integrates a camera, a microphone, a frequency-scalable (up to 624 MHz) CPU, 16 MB FLASH, and 64 MB RAM onto a single device. The device then connects with a standard sensor network mote to form a wireless camera mote. With reasonably low power consumption and extensive algorithmic libraries running on a decent operating system that is easy to program, CITRIC is ideal for research and applications in distributed image and video processing. Its capabilities of in-network image processing also reduce communication requirements, which has been high in other existing camera networks with centralized processing. Furthermore, the mote easily integrates with other low-bandwidth sensor networks via the IEEE 802.15.4 protocol. To justify the utility of CITRIC, we present several representative applications. In particular, concrete research results will be demonstrated in two areas, namely, distributed coverage hole identification and distributed object recognition.

We investigate a practical approach to solving one instantiation of a distributed hypothesis testing problem under severe rate constraints that shows up in a wide variety of applications such as camera calibration, biometric authentication and video hashing: given two distributed continuous-valued random sources, determine if they satisfy a certain Euclidean distance criterion. We show a way to convert the problem from continuous-valued to binary-valued using binarized random projections and obtain rate savings by applying a linear syndrome code. In finding visual correspondences, our approach uses just 49% of the rate of scalar quantization to achieve the same level of retrieval performance. To perform video hashing, our approach requires only a hash rate of 0.0142 bpp to identify corresponding groups of pictures correctly.

The paper describes a target tracking system running on a heterogeneous sensor network (HSN) and presents results gathered from a realistic deployment. The system fuses audio direction of arrival data from mote class devices and object detection measurements from embedded PCs equipped with cameras. The acoustic sensor nodes perform beamforming and measure the energy as a function of the angle. The camera nodes detect moving objects and estimate their angle. The sensor detections are sent to a centralized sensor fusion node via a combination of two wireless networks. The novelty of our system is the unique combination of target tracking methods customized for the application at hand and their implementation on an actual HSN platform.

We consider the problem of establishing visual correspondences in a distributed and rate-efficient fashion by broadcasting compact descriptors. Establishing visual correspondences is a critical task before other vision tasks can be performed in a wireless camera network. We propose the use of coarsely quantized random projections of descriptors to build binary hashes, and use the Hamming distance between binary hashes as the matching criterion. In this work, we derive the analytic relationship of Hamming distance between the binary hashes to Euclidean distance between the original descriptors. We present experimental verification of our result, and show that for the task of finding visual correspondences, sending binary hashes is more rate-efficient than prior approaches.

We present a compressed domain scheme that is able to recognize and localize actions at high speeds. The recognition problem is posed as performing an action video query on a test video sequence. Our method is based on computing motion similarity using compressed domain features which can be extracted with low complexity. We introduce a novel motion correlation measure that takes into account differences in motion directions and magnitudes. Our method is appearance-invariant, requires no prior segmentation, alignment or stabilization, and is able to localize actions in both space and time. We evaluated our method on a benchmark action video database consisting of six actions performed by 25 people under three different scenarios. Our proposed method achieved a classification accuracy of 90%, comparing favorably with existing methods in action classification accuracy, and is able to localize a template video of 80 x 64 pixels with 23 frames in a test video of 368 x 184 pixels with 835 frames in just 11 s, easily outperforming other methods in localization speed. We also perform a systematic investigation of the effects of various encoding options on our proposed approach. In particular, we present results on the compression-classification tradeoff, which would provide valuable insight into jointly designing a system that performs video encoding at the camera front-end and action classification at the processing back-end.

Given a large collection of videos containing activities, we investigate the problem of organizing it in an unsupervised fashion into a hierarchy based on the similarity of actions embedded in the videos. We use spatio-temporal volumes of filtered motion vectors to compute appearance-invariant action similarity measures efficiently - and use these similarity measures in hierarchical agglomerative clustering to organize videos into a hierarchy such that neighboring nodes contain similar actions. This naturally leads to a simple automatic scheme for selecting videos of representative actions (exemplars) from the database and for efficiently indexing the whole database. We compute a performance metric on the hierarchical structure to evaluate goodness of the estimated hierarchy, and show that this metric has potential for predicting the clustering performance of various joining criteria used in building hierarchies. Our results show that perceptually meaningful hierarchies can be constructed based on action similarities with minimal user supervision, while providing favorable clustering performance and retrieval performance.

We present a compressed domain scheme that is able to recognize and localize actions in real-time. The recognition problem is posed as performing a video query on a test video sequence. Our method is based on computing motion similarity using compressed domain features which can be extracted with low complexity. We introduce a novel motion correlation measure that takes into account differences in motion magnitudes. Our method is appearance invariant, requires no prior segmentation, alignment or stabilization, and is able to localize actions in both space and time. We evaluated our method on a large action video database consisting of 6 actions performed by 25 people under 3 different scenarios. Our classification results compare favorably with existing methods at only a fraction of their computational cost

To compare spatial patterns of gene expression, one must analyze a large number of images as current methods are only able to measure a small number of genes at a time. Bringing images of corresponding tissues into alignment is a critical first step in making a meaningful comparative analysis of these spatial patterns. Significant image noise and variability in the shapes make it hard to pick a canonical shape model. In this paper, we address these problems by combining segmentation and unsupervised shape learning algorithms. We first segment images to acquire structures of interest, then jointly align the shapes of these acquired structures using an unsupervised nonparametric maximum likelihood algorithm along the lines of 'congealing' (E. G. Miller et al., 2000), while simultaneously learning the underlying shape model and associated transformations. The learned transformations are applied to corresponding images to bring them into alignment in one step. We demonstrate the results for images of various classes of Drosophila imaginal discs and discuss the methodology used for a quantitative analysis of spatial gene expression patterns.

The correction of bias in magnetic resonance images is an important problem in medical image processing. Most previous approaches have used a maximum likelihood method to increase the likelihood of the pix- els in a single image by adaptively estimating a correction to the unknown image bias field. The pixel likelihoods are defined either in terms of a pre-existing tissue model, or non-parametrically in terms of the image's own pixel values. In both cases, the specific location of a pixel in the im- age is not used to calculate the likelihoods. We suggest a new approach in which we simultaneously eliminate the bias from a set of images of the same anatomy, but from different patients. We use the statistics from the same location across different images, rather than within an image, to eliminate bias fields from all of the images simultaneously. The method builds a "multi-resolution" non-parametric tissue model conditioned on image location while eliminating the bias fields associated with the orig- inal image set. We present experiments on both synthetic and real MR data sets, and present comparisons with other methods.