At Peloton, we constantly strive to improve the member experience by highlighting personalized content that speaks to each individual user. One area of focus is our landing page, the homescreen, consisting of numerous rows of class recommendations used to captivate our users and guide them through our growing catalog of workouts. In this paper, we discuss a strategy we have used to increase the rate of workouts started from our homescreen through a Thompson sampling approach to row ranking. We also explore a potential improvement with a collaborative filtering method based on user similarity calculated from workout history.

At Peloton, we are challenged to not just surface relevant recommendations of fitness classes to our members, but also timely ones. As our fitness content library expands, we continually produce classes on certain themes which are most timely during a narrow time window. To address this challenge, we provide some control over our recommendations to external stakeholders, such as production and marketing teams. They enter timed boosts of certain classes during the windows they are relevant in. We have built out algorithms which take these desired classes and elevate the number of impressions for them, while preserving members’ engagement with our recommendations. In this paper, we discuss the system, the algorithms and some results from a few A/B tests showing how boosting works in practice.

Peloton is at the forefront of the at-home fitness market, with two business pillars: (a) a line of connected fitness equipment, and (b) a subscription-based service that offers access to a rich catalog of high quality fitness classes. As of May 2021, the total member base for Peloton stood at over 5.4 million, who took more than 170 million workouts. Peloton classes cover a diversity of instructors, languages, fitness disciplines, durations, intensity and muscle groups. On the other side, each user has their own specific fitness goals, time available to work out, fitness equipment and level of skill or strength. This diversity of content and individuality of user needs creates the need for a recommender system capable of personalizing the Peloton experience.

Frequent subgraph mining (FSM) has important applications in areas such as bioinformatics, social networks and others. In this paper, we present a highly scalable approach called ParGraph that can efficiently mine from a single graph in both distributed as well as shared-memory based systems. In a distributed environment, we can leverage the local memory of multiple compute nodes for storing a large number of intermediate states for enumerating patterns. To address the skewness in the pattern generation tree, our approach uses a novel hybrid load balancing scheme to efficiently distribute workload across both processes and threads. Our experiments demonstrate good speedups using message passing interface (MPI) and OpenMP threads.

Despite increasing popularity, Location-based Services (LBS) (e.g., searching nearby points-of-interest on map) on mobile handheld devices have been subject to major privacy concerns for users. The existing third-party privacy protection methods hide the exact location of users from service providers by sending cloaking regions (CR) that contain several other user locations in the vicinity. However, this has not ensured LBS full immunity from the privacy concerns. In this paper, we describe a serious privacy problem of LBS called multi-query attack. In this attack, the exact location of the service requester can be inferred by the adversary through obtaining cloaking regions that are shrunk or extended in subsequent queries. This problem can be addressed by judiciously retaining, over a period of time, the cloaking regions for the same set of users. Most methods in the literature are weakened for considering only a static snapshot of users during evaluation. Thus, any update due to user movements in real time becomes very costly. Our proposed approach, ANNC (Adaptive Nearest Neighborhood Cloaking) ,emphasizes developing disjoint sets of users dynamically over time in order to share the common CRs. The CRs are organized in balanced binary trees with restricted height. Thus ANNC achieves the balance between search efficiency and quality of cloaking with higher anonymity levels. The experimental evaluation demonstrates that ANNC will be more efficient in practice than other well-known approaches.

We design and implement a low cost, easy to deploy and lightweight location tracking solution that operates without the need for a fixed infrastructure. We call it "WiFi radar," which uses the signal strength of radio transmissions to determine the position (distance and direction) of devices with near linear approximation. Notwithstanding the challenges that radio frequency signals pose for location determination, the accuracy and precision of our system is relatively high. Our application is user-friendly, customizable and shows graphical as well as list views of the located objects. To the best of our knowledge, our system is the first endeavor that implements a location tracking system without the use of any fixed infrastructure and it may be the cheapest solution built so far.

The emergence of new technologies and the proliferation of mobile and handheld devices have facilitated development of context-based services. Common examples include the location based services. However, the revelation of the context and static information gives rise to new and very complex privacy concerns in such services. In this paper, we propose a formal model to regulate the privacy level of information passed to the service provider without disrupting the quality of information required for service access. The framework takes on two forms based on the receiver of the request generalization task to a trusted party or local device. A formal attack model assumes different levels of knowledge by the attacker and demonstrates the relative concerns of static information over contexts.

Users of handheld devices may want to share services resources with each other in pervasive computing environment. Availability of information regarding users may lead to privacy violations while communicating with others for services and resources. Providing a model to resolve this issue in centralized or distributed environment is not desirable or not feasible if we really want to achieve ubiquity. In this paper, we present a privacy aware model for resource sharing in pervasive computing environment with an illustrative example. Our model exploits the concepts of policies and ordered binary decision diagrams.

The widespread prevalence of pervasive devices and applications has raised the concerns of privacy. Granting Access to Resources and Context sensitive information causes information leakage through inference or obfuscation. Again, the open and dynamic collaborative environment of pervasive computing has rendered the traditional access control models like Role based models to be unfit. Even though Trust based models came to rescue in such circumstances, privacy is mostly compromised in the big picture. In this paper, we have drawn examples from pervasive computing environment and illustrated some scenarios of privacy violation. We have presented a trust based access control model for pervasive healthcare environment to prevent information leakage on accessing constraint information that yields privacy violation in the end. We have addressed information leak from three perspectives of constraint information satisfaction to grant access to the resources. They are satisfy any, satisfy all and hierarchical constraints. Furthermore, the model eliminates requirements for maintaining any keys or access rights certificates for privacy protection. This lightweight model helps ensure the resource constrained small pervasive devices like PDA, cell phones use the access control model effectively and efficiently with privacy of the individuals preserved in the first place.

From container terminals, healthcare services, libraries to household, the role of Asset Tracking Applications has become indispensable. As the organizations grow, so do their volume of assets, soon it becomes very cumbersome to track these assets in real time and accurately prepare a financial report to avoid overbuys. Rapid development of Wireless Area Network and Radio Frequency transmitting active and passive devices like RFID tags have bolstered the deployment of Wireless Asset Tracking applications in all these disciplines. But apart from organizations where costly network infrastructures are in place to support such a reliable asset tracking task, the areas like ports, warehouses, truck stops, parks, mines, rescue spots still suffer from appropriate solutions that can cope with the adverse scenario of being devoid of infrastructure. Even high end Location Based Systems (LBS) like GPS are not scaled well in such situations. To meet these challenges, our approach presents a common platform that can locate different active and passive RF transmitting objects over the range of distances on small handheld devices. To effectively utilize the resources of the constrained handheld devices like PDA and cell phones, a light weight algorithm has been used. Our approach follows an extensible and modular architecture which offers applications from different platforms to customize and extend their functionalities. Smart Tracker has been implemented and evaluated with PDAs, RFID tags, WiFi sources for both indoor and outdoor applications.