Edgelab

Edge AI Test Bed

This project aims to define a distributed software architecture that allows to optimize the distribution of machine learning (ML) inference tasks, in systems characterized by wireless connectivity and heterogeneity in hardware, software, and workloads. By taking advantage of edge-based service mesh mechanisms, we investigate how to efficiently orchestrate the ML inference provisioning between resource-constrained edge nodes and IoT end-devices. Orchestration...

Robotic Intervention Learning at the Edge

A key application of edge and fog computing is that of autonomous driving. However, in order to robustly and safely maneuver around people, current approaches will need to learn at a huge scale. Scalable robot learning from seamless human-robot interaction is critical if robots are to solve a multitude of tasks in the real world. Current approaches to imitation learning...

Coded Computing

The Coded Computing project focuses on addressing variability in response time due to abnormal system behavior, such as shared resources, background daemons, maintenance activities, network bandwidth, queuing, and energy management. Under these system challenges, we explore ways to overcome the performance and scalability problems of MDS-based coded computing. In particular, coded computing based on MDS coding needs to collect at...

Secure Machine learning

Most machine learning (ML) systems are designed with an assumption that test inputs will belong to a particular pre-defined training distribution. However, most real-world ML applications, such as autonomous driving, biometric authentication are likely to receive inputs beyond the training distribution. This is where Open-wold machine learning becomes necessary. It not only learns that task on the training distribution but...

Massive Machine-Type Communications

We propose UL random access protocols using pseudo-random beamforming. Specifically, for 1) infinite backlog systems, uniform loads, and channel conditions, and 2) finite backlog systems, uneven loads, and channel conditions, we develop joint beam selection and random access schemes that can provably achieve fractions of the optimal capacity region with low overhead regardless of the number of interfering devices.

Big Data Sharing

The Sharing Big Data project seeks to understand the impact of the data sharing on the downstream market prices and mobile users' economic welfare. If the MNO and MVNO have the same contextual data, both operators may be able to provide users a similar QoS for edge services, leading to fierce competition between operators and reduced profits. Thus, understanding how...

Hurts to Be Too Early: Benefits and Drawbacks of Communication in Multi-Agent Learning.

In this work, we study the problem of multi-agent reinforcement learning in cooperative environments, and analytically evaluate the effects of information sharing on both the coordination and learning of the agents. We are particularly interested in the role of communication when agents have heterogeneous capabilities in assessing their shared environment. This is motivated by the possible heterogeneity in agents’ platforms;...

Personalized Thread Recommendations for MOOC Forums

Thread recommendations that are based on timing, content, and latent user interest are more effective than methods that ignore any of those aspects.

Shred and Spread

Public cloud storage has recently become popular among individuals storing personal files and enterprises sharing business documents. However, cloud service providers (CSPs) offer fairly rigid services, which cannot be integrated to meet individual users’ needs. In this work, we provide a client-defined cloud storage service that integrates multiple autonomous CSPs into one unified cloud and allows individual clients to specify...

Networked Drone Cameras

A network of drone cameras can be deployed to cover live events, such as high-action sports game played on a large field, but managing networked drone cameras in real-time is challenging. Distributed approaches yield suboptimal solutions from lack of coordination but coordination with a centralized controller incurs round-trip latencies of several hundreds of milliseconds over a wireless channel. We propose...

Social Learning Efficiency

A Social Learning Network, or SLN, encapsulates a range of scenarios in which a number of people learn from one another through structured interaction. These are networks that exist between learners, instructors, and also modules of information. In this work, we ask: How efficient are real-world SLN, and how can their efficiencies be improved? To answer these questions, we have...

Securing Tor

The Tor network is a widely used system for anonymous communication. However, Tor is known to be vulnerable to attackers who can observe traffic at both ends of the communication path. We present a suite of new attacks, called Raptor, that can be launched by Autonomous Systems (ASes) to compromise user anonymity.

Cloud Virtual Service Provider

Cloud service providers (CSPs) often face highly dynamic user demands for their resources, which can make it difficult for them to maintain consistent quality-of-service. Some CSPs try to stabilize user demands by offering sustained-use discounts to jobs that consume more instance-hours per month. These discounts present an opportunity for users to pool their usage together into a single "job." In...

Clouds with Deadlines

There exist many different modern approaches to resource allocation in computing clusters, ranging from the simplistic (FIFO, highest priority first, and earliest deadline first) to the more complex (fair scheduling, etc). However, many of these algorithms focus on maximizing optimization problems that have very basic job utilities that are either constant or binary with respect to a deadline time. We...

Client Control of HetNets

The theory for HetNet RAT Selection is not limited to only 3G, LTE and 802.11 networks. In collaboration with teams under Dr. Caire and Molisch at USC, we at the EDGE Lab are investigating what makes millimeter-wave radio unique from other types of RATs, and how mmWave RATs can be characterized in terms of HetNet RAT Selection.

Multi-resource fairness

People have been arguing about fairness for thousands of years, including academics working in economics, politics, philosophy, and engineering. Though fairness can consequently mean many different things, we focus on a specific aspect of fairness: defining the fairness of an allocation of resources to different people. We use a mathematical theory of fairness that is based on four fairness axioms,...

LTE Multicast

With recent standardization and deployment of LTE eMBMS, cellular multicast is gaining traction as a method of efficiently using wireless spectrum to deliver large amounts of multimedia data to multiple cell sites. Content delivery over cellular networks has been a challenge, with the last hop RAN being the bottleneck and a significant source of latency. CDNs relieve the backbone bandwidth...

Completion-time-aware Scheduler

Imagine that car rental companies such as Hertz rent out cars to their customers without knowing when they are going to return them. It is impossible for the companies to know how long the next customer has to wait until there is a car available for him/her, let along guaranteeing the quality of service (QoS) received by the customers. Similarly,...

Cloud Bidding

Amazon’s Elastic Compute Cloud (EC2) uses auction-based spot pricing to sell spare capacity, allowing users to bid for cloud resources at a highly reduced rate. Amazon sets the spot price dynamically and accepts user bids above this price. Jobs with lower bids (including those already running) are interrupted and must wait for a lower spot price before resuming.

Time-dependent Pricing

Datami is motivated by the growth in mobile (and wired) demand for data and ISPs' (Internet Service Providers) increasing inability to meet this demand. To make it worse, the heavy usage concentrates on several peak hours in a day, forcing ISPs to overprovision according to that. Even charging by monthly usage overages, as AT&T and Verizon have started doing in...

Quota-aware Video Adaptation

Two emerging trends of Internet applications, video traffic becoming dominant and usage-based pricing plans becoming prevalent, are at odds with each other. On one hand, videos, especially on high-resolution devices (e.g., iPhone 5, iPad, Android tablets), consume much more data than other types of traffic; for instance, 15 min of low bitrate YouTube videos per day uses 1 GB a...

Traded Data Plans

The growing volume of mobile data traffic has led many Internet service providers (ISPs) to cap their users' monthly data usage, with steep overage fees for exceeding their caps. In this work, we examine a secondary data market in which users can buy and sell leftover data caps from each other.

Engagement Modeling

Current approaches to learning analytics are focused mainly on providing feedback to learners about their knowledge states, based on their responses to assessment questions. Accounting for additional cognitive factors (most importantly, learner engagement) has the potential to yield more effective learning analytics and feedback. However, measuring these factors has remained a difficult task. Recently, the emergence of online learning platforms,...

Individualization

One of the major pain points in online learning today is the scale-efficacy tradeoff of learning. The best example of this is perhaps Massive Open Online Courses (MOOCs), which have showcased orders of magnitude larger student bodies than traditional classrooms but with completion rates that hover in the single digits. The problem is that in settings where the teacher to...

Clickstream Analytics

Learning technology platforms today can be equipped with infrastructure to capture fine-granular behavioral data about students as they proceed through a course. This includes, for example, the sequence of clicks made while watching a video lecture, answering a quiz question, or posting in a discussion. The resultant “big” data presents unprecedented opportunities to study the process by which learning occurs.

< >

Chiang receives State Department Appointment

On Dec. 16, 2019 Mung Chiang began a one year leave of absence from his role as the Dean of Engineering at Purdue to serve as the director of the Office of the Science and Technology Adviser to the Secretary of State (STAS). The role draws on his expertise to advise the U.S. Department of State on the economic security of the nation’s global technology initiatives, such as artificial intelligence, 5G wireless networks, energy infrastructure, cybersecurity for autonomous systems, and trusted microelectronic chips. https://www.state.gov/biographies/mung-chiang/

Princeton Edge Lab 10th Conference

On May 17, 2019, an exciting conference will be held at Princeton on edge/fog computing: program and registration at http://www.edge10.princeton.edu

Chiang becomes Purdue Dean

On July 1, 2017, Mung Chiang joined Purdue University as the John A. Edwardson Dean of the College of Engineering. Purdue Engineering is one of the largest in the nation and has always ranked in the top 10. He is also the Roscoe H. George Professor of Electrical and Computer Engineering at Purdue and maintains a visiting appointment at Princeton.