Colloquium

Mesh or Multiple-Tree: A Comparative Study of Live P2P Streaming Approaches

Existing approaches to P2P streaming can be divided into two general classes: (i) tree-based approaches, use push-based content delivery over multiple tree-shaped overlays, and mesh-based approaches use swarming content delivery over a randomly connected mesh. Previous studies have often focused on a particular P2P streaming mechanism and no comparison between these two classes has been conducted. In this talk, we compare and contrast the performance of representative protocols from each class using simulations. We identify the similarities and differences between these two approaches.

Classifying Hosts Based on the Destination Address Distribution of their Outgoing Network Connections

The increased prevalence of open wireless networks and portable computers has reduced network administrators control over, and knowledge of, the machines in their networks. Efficient means of classifying these hosts are required so that administrators can make informed decisions about how they manage their networks. One effective way of making broad classifications about machines is by monitoring their network behavior, and specifically the distribution of the destination addresses they connect to.

Vulnerabilities and Opportunities in SMS-Capable Cellular Networks

Cellular networks are a critical component of the economic and social infrastructures in which we live. In addition to voice services, these networks deliver alphanumeric text messages to the vast majority of wireless subscribers. To encourage the expansion of this new service, telecommunications companies offer connections between their networks and the Internet. The ramifications of such connections, however, have not been fully recognized. In this talk, we evaluate the security impact of the SMS interface on the availability of the cellular phone network.

Network coding as an efficient scheduling algorithm for large scale systems

Designing large scale content distribution systems, such as peer-to-peer networks, is a very challenging task. One fundamental problem is to schedule optimally (e.g. as to minimize the time to distribute the content to all users) using local information only. Using ideas borrowed from network coding theory, in particular by treating data as algebraic entities that can be transformed, we have designed systems that effectively use the resources of the network and, yet, do not use global information.

On the State of Spoofing Prevention

IP source address spoofing has plagued the Internet for many years. Attackers attempt to spoof source addresses in order to mount attacks and redirect blame. Researchers have proposed many mechanisms to combat and prevent spoofing, with varying levels of success. With the prevention mechanisms available today, where do we stand? How much of a problem is spoofing today? How do the various prevention mechanisms compare? This study looks into the current state of IP spoofing and IP spoofing prevention.

AutoDL: Automated Neural Network Architecture Search for Open & Inclusive AI

AI is delivering significant impact to human society. In areas such as game playing, image classification, and speech recognition, AI algorithms may have already surpassed human experts' capability. We are observing transformations AI produces to industry sectors such as social media, finance, and transportation. My talk covers two parts: (1) an overview of Baidu Brain and (2) a brief introduction to a new initiative called Open & Inclusive AI at Baidu where we use deep learning to design deep learning networks.

Directed Research Project Presentations

The first milestone in the CIS Ph.D. program is the Directed Research Project (DRP). Each student must complete this milestone within their first two years in the program. Typically, students devote the summer after their first year in the program exclusively to their research projects. In this colloquium, seven of our Department's Ph.D. students will present summaries on the progress on their DRP's over this past summer.

Accelerate Parallelism in Large Scale Machine Learning

Parallelism is the key strategy to improve the efficiency for solving large scale machine learning tasks, by involving multiple workers to work in parallel. In this talk, three frameworks are introduced to accelerate the parallel computation in minimizing ML objectives. The first framework is to study the asynchronous parallelism to avoid the synchronization overhead in the traditional syntonization framework.

Pages

Subscribe to RSS - Colloquium