Fan Ye
   Associate Professor
   Electrical and Computer Engineering
   Stony Brook University
   Stony Brook, NY 11794

   Phone : +1 631 632 8393

   Office : Light Engineering 217
   Email: (at) stonybrook DOT edu

New: updates to 2017 publications

I'm an Associate Professor in the Electrical and Computer Engineering Department at Stony Brook University, also affiliated with the Smart Energy Techologies (SET Cluster). Before that I worked at IBM T.J. Watson Research Center as a Research Staff Member since 2004.

My research interests are in mobile and embedded sensing systems and applications (e.g., location based services, healthcare), edge computing, Internet-of-Things, sensor networks, energy and power efficiency. I have over 90 referred publications which in total have been cited more than 10000 times according to Google Scholar, and 26 granted/pending multi-national/US patents/applications.

I have worked on various projects in these areas: sensing infrastructure for aging in place, enterprise-scale IoT management, indoor floor map construction, indoor vehicle/smartphone tracking and localization, mobile crowdsensing and its infrastructure, cloud based publish/subscribe service, wide area messaging, high availability and collaborative stream processing systems, sensor networks (robust data delivery, sensing coverage, and false data injection attacks).

Some of my recent projects include:

Location Based Services (LBS)

I have multiple projects addressing scalable and accurate indoor map construction, and indoor localization/tracking.

BatTracker: Accurate, infrastructure-free 3D tracking of mobile devices without any external hardware. The paper will appear in ACM SenSys 2017.

BatMapper: We leverage acoustic signals to create floor maps in just a few minutes, with accuracy comparable to the state of the art, at two orders of magnitude less data and efforts. The paper appeared in ACM Mobisys 2017 with a short video

Knitter: a single, novice user can create accurate floor maps in just one hour, cutting down data amounts and user efforts by one order of magnitude compared to prior work. The paper appeared in IEEE Infocom 2017.

Jigsaw: We leverage crowdsensed data from mobile users to construct complete and accurate floor plans automatically. It's the first to create high accuracy and complete maps using commodity mobile devices. The paper appeared in ACM Mobicom 2014.

Garage Mapping: given vehicles' fragmented movement trajectories, we develop several algorithms to piece together them for complete garage maps. The paper appeared in IEEE ICC 2016.

VeTrack: we track the real time movements of vehicles in uninstrumented environments (e.g., underground parking structures) where no radio signal (e.g., WiFi, GPS, cellular) is available. The paper appeared in ACM Sensys 2015. A video demo shows the real time tracking in a garage.

Sextant: instead of WiFi signals, we leverage abundant physical objects as reference to triangulate the user location. The papers appeared in Infocom 2014, ICC 2014 and TMC.

Peer acoustic assisted localization: we exploit the more accurate acoustic ranging among peer phones to pose additional constraints on the location of a phone for much higher accuracy (~1m) compared to WiFi only methods. We have built a prototype and a demo video is available. The paper appeared in Mobicom 2012 and TMC


Pervasive Edge Computing

We are developing a new mobile sensing/computing paradigm where edge devices (smartphones, vehicles, IoT) communicate with each other directly to discovery data available in the surrounding environments. They exchange, process data spontaneously using data centric primitives.

Pervasive Data Sharing: agile, light-weight peer data discovery and retrieval among opportunitically congregated edge devices, specifically addressing short contact durations and uncertain data availability. The paper appeared in ICDCS 2017.

Fair Data Caching: data caching among edge devices of heterogeneous ownership must achieve fairness to ensure and encourage participation. The paper appeared in ICDCS 2017.


Smart Environment

We are developing a holistic hardware/software architecture for fine grained access, command execution assurances and scalable management for enterprise Internet-of-Things (e.g., lights, door locks, window blinds, A/C etc. in tens of campus buildings of a university). There are potentially hundreds of thousands of smart objects accessed by tens of thousands of people taking different roles and responsibilities.


Rechargeable Sensor Networks

Using autonomous vehicles equipped with batteries, solar panels and wireless charging capabilities, we can replenish the energy of sensor nodes to achieve perpetual operation.

Netwrap : use named data based methods to collect recharging requests and schedule the vehicle's charging activities.

Multi-vehicle coordination: coordinate the charging schedules of multiple vehicles for optimal network operation.

Vehicle movement costs and capacity: energy costs in vehicle movements and battery capacity are considered when deciding the vehicle's charging schedule.

Resonant repeaters: a vehicle can recharge multiple sensor nodes at once when equipped with resonant repeaters.

Solar energy harvesting: combine the complementary strengths of solar panels and wireless charging capability.



Earlier work on mobile sensing, stream processing and messaging systems

Smartphone privacy reveals that mobile advertisers can easily infer users' social connections with a few weeks's scavenging of private data.

Mobile Crowd Sourced Sensing leverages large numbers of mobile devices with data collecting capabilities to provide input for Smarter Planet applications. It seeks to build a common infrastructure that interacts with devices, configures their data gathering activities based on declarative requirements from backend applications

Crowdedness Detection allows mobile devices to quickly and efficiently identify most of the neighbors by leveraging the knowledge of each other. It provides a basic function that enables interactive user experience in many Mobile Crowdsensing applications.

Cloud based publish/subscribe service exploits the natural skewness in data distribution to provide a scalable and elastic attribute-based pub/sub service that has linear capacity scaling and multi-fold higher throughput than existing systems.

Responsive, Reliable and Real-Time Messaging delivers critical and time-sensitive information across wide area networks among end points in physical and digital worlds. It adopts multiple techniques to discover reliable routing paths and schedule message deliveries in time.

Hybrid high availability addresses "transient unavailability" in stream processing systems by intelligently switching between active and passive standby mechanisms. It reduces recovery time by two-thirds and message overhead by 80%, and reproduces results as if no failure had happened.

Cooperative Stream Processing enables multiple autonomous stream processing systems to collaborate and further scale up the processing capabilities. It enables analysis and processing that are difficult or impossible in individual systems. 

Federated Resource Discovery provides a unified search interface so that stream processing applications can locate and utilize resources in different autonomous systems over wide area networks.


Negotiation Management decides the optimal schedule for reserving remote resources among stream processing systems, such that an application execution plan can be satisfied within the required budget and time constraints.



Even earlier work on wireless sensor networks

On robust, large scale data forwarding, energy efficiency and security, some of which are highly visible:

GRAdient Broadcast (GRAB) utilizes a novel concept of cost-field to enable robust and scalable data forwarding under extremely adverse conditions.

Two-tier Data Dissemination (TTDD)  was the first protocol that delivers data efficiently to multiple users that are in constant motion by building a virtual infrastructure.

Probe Environment and Adaptive Sleeping (PEAS) extends system lifetime in proportion to nodes' deployment density by maintaining a certain distance between working neighbors.

Statistical En-route Filtering (SEF)  was the first proposal that detects bogus data injected by compromised insider nodes. It filters such data en-route, thus avoiding delivering false information or wasting energy.

Location Dependent Keys (LDK) provides secure report generation against large numbers of compromised nodes. By binding the keys to a node's location rather than its identity, the user can verify whether a report does originate from the claimed location.

Probabilistic Nested Marking for Traceback identifies the exact origin of attacking packets, such that compromised insider nodes can be isolated from the network. It was the first work that addresses colluding attacks by multiple compromised nodes.


Email me if you have any questions about my work.