Jarrell Waggoner, will show us how to build an extension for the Chrome browser. Jarrell Waggoner is a PhD student in our department. He implemented the nonpartisan me which received some attention. This talk is part of CSCE 242 but is open to all students.

In this interactive walk-through, we will create a custom application from scratch using proven Microsoft tools and technologies and a healthy dose of new approaches to software development. This session will demonstrate how easy it can be to create rich single page applications and robust client experiences while still leveraging .NET languages and tools to define the bulk of the business logic and processes (not just a tangled mess of JavaScript). While this will be a high-level demonstration, it still show how to handle hard-core problems like symmetric client and server-side template rendering, complex custom validation and event-driven model manipulation using both client and server-side rules. Though the problems solved will be hard, the solutions demonstrated will be both elegant and easy to understand. This is an invited talk by Jamie Thomas. This talk is part of CSCE 242 but is open to all students. Jamie Thomas is the Director of Software Development at VC3, an IT and software services company, headquartered here in downtown Columbia, in the IT-ology building.

Appathon contestants will show off their submissions, free lunch and t-shirts will be provided, and prizes will be awarded! There will also be an introductory workshop in android development. So come out and get some free food, some sweet swag, and see the cool stuff your fellow students have created. See event details.

MS Thesis Defense: Gary Fredericks We study a family of network formation models to determine how payment rules affect the final network topologies that emerge. In our model a set of nodes starts out without any edges and the nodes must pay for the creation of edges using one of several different payment mechanisms, for example: one node pays for the whole edge, the cost is shared equally between the two nodes, etc. We show how the set of networks formed by some payment rules are subsets of those formed by other rules. We also perform extensive empirical tests on networks of up to 10 nodes. These tests reveal some interesting patterns in the connectivity, stability, and fairness of the networks generated by the various payment rules given a fixed link cost.

PhD Defense: Zhenhua Liu The open nature of wireless communication makes it easy for adversaries to either inject random signals to wireless channels harming the network availability, or eavesdrop on the communication breaching the location privacy. Those threats are challenging to cope with, because most of them cannot be addressed by traditional cryptographic methods. One of the attacks that can easily imperil the availability of networks is jamming attacks. An adversary can launch a jamming attack either by bypassing MAC-layer protocols and keeping sending packets, or by emitting radio signals to a particular channel. To cope with jamming attacks, in this dissertation, we focus on developing mechanisms to localize jammers. We examine how jammers affect networks in terms of signal strength, nodes’ communication range, and network topologies, and present how to measure these jamming effects. To localize a jammer, we design an Adaptive Least-Squares-based (LSQ-based) algorithm which performs localization by exploiting the changes of communication range. Then, to further improve the localization accuracy, we propose an error minimizing framework that can localize not only one but also multiple jammers through utilizing the strength of jamming signals (JSS). To evaluate the effectiveness of our proposed localization schemes, we conducted real-world experiments using a testbed of MicaZ, and then carried out extensive performance studies in large-scale networks by simulation. Another problem that cannot be solved by traditional cryptographic method is the location privacy issue. We study this issue in wireless sensor network because the locations of the sink nodes are critically important to the viability of wireless networks, and such information can be easily determined by attackers. For instance, attackers can eavesdrop on the network communication at several spots and trace back to the sink nodes. Then, they can destroy the sink nodes physically to disable the data collection or dissemination. In this dissertation, we examine the sink location privacy problem from both the attack and defense sides. On the attack side, we present two types of Zeroing-In attacks which allow attackers to identify the sink location by estimating the hop count or the arrival time of a broadcast packet at a few spots in the network. To cope with the Zeroing-In attacks, we propose a directed-walk-based scheme and validate that it is effective in deceiving adversaries at modest energy costs."

Dr. Jason O'Kane will be doing a Talkback at the Nickelodeon Theatre.

PhD Defense - Mikhail Simin In study of diseases and their molecular foundation and evolution, it is critical to study the three dimensional structure of biological macromolecules. Structural characterization of biological macromolecules is further motivated by the fact that biomolecules with defined function(s) exhibit a correlation between their structure and their function. One functional group of bio¬logical macromolecules is proteins: strands of amino acids that form various shapes, and perform various cellular functions. Determining the three dimensional structure of a protein becomes a pivotal point in protein analysis and medicinal studies. Proteins are polymers that are composed of 20 fundamental units of Amino acids. Amino acids vary primarily through their side chain, while sharing nearly identical backbone atoms. In several instances studying the backbone structure of a protein is of significant benefit, as opposed to the all-atom study. It has also been shown that given a protein backbone the side-chains can be places analytically or computationally. One of the advantages of backbone-only study is that obtaining experimental data, such as residual dipolar coupling, is significantly easier than data for side-chains. In the recent years protein structure determination has been assisted by residual dipolar coupling (RDC) data. RDCs show promise as a powerful source for structure determination, not only due to their sensitivity, but also their applicability in macromolecules such as large proteins, membrane-anchored proteins, homo-multimeric protein complexes, carbohydrates, and nucleic acids. Although RDCs are commonly used with a minimum contribution in structure refinement of proteins, their information content extends to de novo structure elucidation. RDCs can be collected fairly easily yet common computational tools do not take maximum advantage of these data. A single RDC datum significantly restrains the possible orientation of a pair of interacting nuclei within a protein; this gives grounds for exploration of minimal data requirements for structure determination. If RDC data are utilized to their maximum potential they can become an informative means of structure determination. This work presents REDCRAFT software package and its advancements in computational structure determination from RDCs. Detailed analyses of the software, and its performance will be presented and discussed in this document. Multiple improvements, as well as new additions to the preceding version of this software will be discussed. In particular, a novel algorithm is presented for solution space decimation addressing REDCRAFT’s native style of search depth selection.

A quick workshop in Android mobile app development led by Drew Heavner (bring your laptop!). Also, pizza! See Event Page.

PhD Defense: Yan Zhang Frequent Itemset Mining (FIM) is a data-mining task that is used to find frequently occurring subsets amongst a database of itemsets. FIM is used in many applications, including those in machine learning and computational biology. In this work, we explore novel hardware architectures to accelerate FIM using multiple Field Programmable Gate Arrays as application-specific coprocessors. In general, FIM is a challenging application to accelerate because it is a data intensive computation and its performance is limited by the available memory bandwidth, and previous work in this area has yielded disappointing results. We develop an efficient hardware accelerator based on ECLAT algorithm. Besides, our approach offers three key advantages to previous efforts. First, we achieve high scalability by dynamically scheduling tasks onto multiple accelerators. Second, we developed a compression scheme for intermediate results and store them onto an on-chip scratchpad memory, significantly reducing the number of off-chip memory accesses. Third, we developed a second data compression scheme for the input data to reduce the total volume of data exchanged over the off-chip memory interface. This compression scheme leverages the bitvector data representation by using a lossless logic minimization-based compression technique that makes single-cycle decoding possible using a novel hardware decoder. Our FPGA coprocessor achieves 29-38 X speedup compared to an optimized x86 implementation. Intermediate compression on scratchpad achieves an additional 2 X speedup, and source bitvector compression achieves an additionally 20-30% speedup.