Matrix factorization (MF) is one of the most popular techniques for product recommendation, but is known to suffer from serious cold-start problems. Item cold-start problems are particularly acute in settings such as Tweet recommendation where new items arrive continuously. In this paper, we present a meta-learning strategy to address item cold-start when new items arrive continuously. We propose two deep neural network architectures that implement our meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted. We evaluate our techniques on the real-world problem of Tweet recommendation. On production data at Twitter, we demonstrate that our proposed techniques significantly beat the MF baseline and also outperform production models for Tweet recommendation.

This demo will showcase LifeJoin, a system that collects raw sensor data from phones and laptop computers to generate interesting events. Given the raw sensor data, LifeJoin implements a number of activity recognition algorithms to generate higher-level events. Furthermore, it uses supervised learning techniques to learn from users' feedback to generate only events of interest. In this demo, the audience will get to interact with the LifeJoin system and be able to examine the internals of LifeJoin.

Modern smartphones are equipped with a wide variety of sensors including GPS, WiFi and cellular radios capable of positioning, accelerometers, magnetic compasses and gyroscopes, light and proximity sensors, and cameras. These sensors have made smartphones an attractive platform for collaborative sensing (aka crowdsourcing) applications where phones cooperatively collect sensor data to perform various tasks. Researchers and mobile application developers have developed a wide variety of such applications. Examples of such systems include BikeTastic [4] and BikeNet [1] which allow bicyclists to collaboratively map and visualize biking trails, SoundSense [3] for collecting and analyzing microphone data, iCartel [2] which crowdsources driving tracks from users to monitor road traffic in real time, and Transitgenie [5], which cooperatively tracks buses and trains.

Real-time transit tracking is gaining popularity as a means for transit agencies to improve the rider experience. However, many transit agencies lack either the funding or initiative to provide such tracking services. In this paper, we describe a crowd-sourced alternative to official transit tracking, which we call cooperative transit tracking.

Traffic delays and congestion are a major source of inefficiency, wasted fuel, and commuter frustration. Measuring and localizing these delays, and routing users around them, is an important step towards reducing the time people spend stuck in traffic. As others have noted, the proliferation of commodity smartphones that can provide location estimates using a variety of sensors---GPS, WiFi, and/or cellular triangulation---opens up the attractive possibility of using position samples from drivers' phones to monitor traffic delays at a fine spatiotemporal granularity. This paper presents VTrack, a system for travel time estimation using this sensor data that addresses two key challenges: energy consumption and sensor unreliability. While GPS provides highly accurate location estimates, it has several limitations: some phones don't have GPS at all, the GPS sensor doesn't work in "urban canyons" (tall buildings and tunnels) or when the phone is inside a pocket, and the GPS on many phones is power-hungry and drains the battery quickly. In these cases, VTrack can use alternative, less energy-hungry but noisier sensors like WiFi to estimate both a user's trajectory and travel time along the route. VTrack uses a hidden Markov model (HMM)-based map matching scheme and travel time estimation method that interpolates sparse data to identify the most probable road segments driven by the user and to attribute travel times to those segments. We present experimental results from real drive data and WiFi access point sightings gathered from a deployment on several cars. We show that VTrack can tolerate significant noise and outages in these location estimates, and still successfully identify delay-prone segments, and provide accurate enough delays for delay-aware routing algorithms. We also study the best sampling strategies for WiFi and GPS sensors for different energy cost regimes.

Sensors capable of sensing phenomena at high data rates on the order of tens to hundreds of thousands of samples per second are now widely deployed in many industrial, civil engineering, scientific, networking, and medical applications. In aggregate, these sensors easily generate several million samples per second that must be processed within milliseconds or seconds. The computation required includes both signal processing and event stream processing. XStream is a stream processing system for such applications. XStream introduces a new data type, the signal segment, which allows applications to manipulate isochronous (regularly spaced in time) collections of sensor samples more conveniently and efficiently than the asynchronous representation used in previous work. XStream includes a memory manager and scheduler optimizations tuned for processing signal segments at high speeds. In benchmark comparisons, we show that XStream outperforms a leading commercial stream processing system by more than three orders of magnitude. On one application, the commercial system processed 72.7 Ksamples/sec, while XStream processed 97.6 Msamples/sec.

Many scientific, financial, data mining and sensor network applications need to work with continuous, rather than discrete data e.g., temperature as a function of location, or stock prices or vehicle trajectories as a function of time. Querying raw or discrete data is unsatisfactory for these applications -- e.g., in a sensor network, it is necessary to interpolate sensor readings to predict values at locations where sensors are not deployed. In other situations, raw data can be inaccurate owing to measurement errors, and it is useful to fit continuous functions to raw data and query the functions, rather than raw data itself -- e.g., fitting a smooth curve to noisy sensor readings, or a smooth trajectory to GPS data containing gaps or outliers. Existing databases do not support storing or querying continuous functions, short of brute-force discretization of functions into a collection of tuples. We present FunctionDB, a novel database system that treats mathematical functions as first-class citizens that can be queried like traditional relations. The key contribution of FunctionDB is an efficient and accurate algebraic query processor - for the broad class of multi-variable polynomial functions, FunctionDB executes queries directly on the algebraic representation of functions without materializing them into discrete points, using symbolic operations: zero finding, variable substitution, and integration. Even when closed form solutions are intractable, FunctionDB leverages symbolic approximation operations to improve performance. We evaluate FunctionDB on real data sets from a temperature sensor network, and on traffic traces from Boston roads. We show that operating in the functional domain has substantial advantages in terms of accuracy (15-30%) and up to order of magnitude (10x-100x) performance wins over existing approaches that represent models as discrete collections of points.

WaveScope is a data management and continuous sensor data system that integrates relational database and signal processing operations into a single system. WaveScope is motivated by a large number of signal-oriented streaming sensor applications, such as: preventive maintenance of industrial equipment; detection of fractures and ruptures in various structures; in situ animal behavior studies using acoustic sensing; network traffic analysis; and medical applications such as anomaly detection in EKGs. These target applications use a variety of embedded sensors, each sampling at fine resolution and producing data at high rates ranging from hundreds to hundreds of thousands of samples per second. Though there has been some work on applications in the sensor network community that do this kind of signal processing (for example, shooter localization [5], industrial equipment monitoring [4], and urban infrastructure monitoring [2]), these applications are typically custombuilt and do not provide reusable high-level programming framework suitable for easily building new signal processing applications with similar functionality. This poster shows how WaveScope supports these types of application in a single, unified framework, providing both high run-time performance and easy application development.