Program Schedule: PhD Forum 4-6 and New Investigators 3-4
Wednesday, October 1st
11:15 a.m. — 12:15 p.m.
PhD Forum 4 - Artificial Intelligence & Learning Systems
Location: Crestone Peak II-IV
Using Planning Techniques to Build a Better World
Presenter: Liangrong Yi (University of Kentucky)
My research area is decision-theoretic planning. I am working on Markov decision process based planning: algorithm design and applications. The first application is decision support for welfare case managers. For that project, I created two new planning algorithms: a receding-horizon planner and a concurrent-action planner. The second application is music, specifically harmony generation. Given the soprano line, my algorithm automatically generates the other three voices of a four-part harmony.
Class Noise Detection through Instance Weighting
Presenter: Umaa D Rebbapragada (Tufts University)
My thesis explores a novel solution to the problem of detecting and eliminating mislabeled training data. Past solutions either discard or correct suspected mislabelings. My approach weights each instance according to its probability of cleanliness. Our hypothesis is that instance weighting improves any supervised method whose performance is impacted by mislabeled training data, including active and semi-supervised learning. Thus, this research has important consequences for the field of supervised learning.
PhD Forum 5 - Security & Privacy
Location: Torreys Peak II
Privacy Preserving Distributed Data Mining: A Game-Theoretic Approach
Presenter: Kamalika Das (University of Maryland, Baltimore County)
This research aims at formalizing a new approach toward privacy preserving data mining in the light of economic game theory. The focus of this research is two-fold: (i) develop a new model of privacy for heterogeneous multi-party distributed data mining environments and (ii) design mechanisms to get rid of some not-so-practical assumptions present in the privacy preserving data mining literature for computation primitives such as sum and inner product.
Intrusion Detection in Wireless Ad hoc Networks Using Cross-layer Designs
Presenter: Geethapriya Thamilarasu (Univeristy at Buffalo)
The goal of our research is to develop an efficient intrusion detection system for wireless ad hoc networks to protect and defend these networks from various security threats and vulnerabilities. We adopt cross-layer interaction techniques to gather relevant audit information from different protocol layers to identify malicious network behaviors. Our research seeks to facilitate the deployment of IDS tool for various wireless networking applications ranging from military to commercial domains.
A Verification Framework for Broken Access Control Attack in Dynamic Web Applications
Presenter: Manar Hasan Aalfi (Queens University)
This thesis will propose a security analysis framework for dynamic web applications. A reverse engineering process will be performed over a dynamic web application to extract a role-based access control security model. A formal analysis will be applied on the recovered model to check access control security properties. This framework may be used to verify if a dynamic web application conforms to the access control polices specified by a security engineer.
PhD Forum 6 - Information Management
Location: Torreys Peak III
Framework for Interactive Massive Volume Visualization
Presenter: Susan Frank (Stony Brook University)
A framework for distributed volume visualization is presented. Out-of-core region growing is introduced for segmentation. Flex-blocks are cells containing empty space and a cropped subvolume, which are used for data reduction. Brick grouping dynamic programming (DP) is used to partition the scene from a DAG of bricks. Moving walls DP uses slab-projection slices, which encrypt empty space information. The cell-tree is a concise representation of ray-traversal dependencies used for ray-task scheduling.
Reliability and Scalability of Large Scale Archival Storage Systems
Presenter: Deepavali Bhagwat (University of California, Santa Cruz)
The design of large scale archival storage systems is a challenging problem. Archival Systems need to be cost effective: storage costs must be reduced by removing data redundancies; they must be reliable: data must be protected to survive storage media failures; they must be scalable: the sheer volume of data required to be preserved makes this imperative. My PhD thesis focuses on these three areas of archival storage systems.
Scalable Content-Based Music Retrieval on Acoustic Datasets via Hashing
Presenter: Yi Yu (Nara Women’s University)
A less emphasized but crucial aspect of content-based music retrieval is how to represent complex audio features to make them easily indexable for quickening the matching of audio data. We address this important issue from two points: i) refining music representation and ii) organizing music documents by index-based approximate techniques. Experimental evaluations prove that the proposed query-by-content music retrieval techniques can be effectively and practically applied in a large audio dataset.
New Investigators 3 - Large Scale Patterns & Parallel Computing
Location: Torreys Peak IV
Large-scale Distributed Storage Systems
Presenter: Aram Shahinfard (University of California, San Diego)
A major challenge in building large-scale distributed applications is designing a storage system that scales to massive volume of data, with a high level of availability and performance. Most such systems run on commodity hardware with lower performance and reliability than high-end servers. Also commercial databases are unable to scale to these requirements. In this paper we present Google’s Bigtable and Amazon’s Dynamo, two custom designed distributed storage systems.
Managing Trust and Interoperability in a Grid
Presenters: Shashi Bhanwar (Thapar University), Seema Bawa (Thapar University)
Security is a major issue that must be resolved in order for the potential of the grid to be fully exploited. In this paper, we discuss architectural, infrastructural and management issues related to grid security. We elaborate Grid Security requirements such as authentication, authorization and confidentiality. We have discussed two major security challenges faced by grid today for its deployment on massive scale and at enterprise level: interoperability and trust management.
FraSPA: A Framework for Synthesizing Parallel Applications
Presenter:Ritu Arora (University of Alabama at Birmingham)
This paper introduces a Framework for Synthesizing Parallel Applications (FraSPA) for multiple-platforms in a user-guided fashion. This research work is motivated by the complexities associated with the Message Passing Interface (MPI) [1], the widely used parallel programming standard. FraSPA will address these complexities and will facilitate the synthesis of parallel programs from sequential application and middleware components. A high-level design approach and initial work to demonstrate the feasibility of FraSPA are presented in this paper.
New Investigators 4 - Algorithms & Applications
Location: Crestone Peak I
Kernel Method for Predicting DNA-Binding Proteins in Yeast from Heterogeneous Data Sources
Presenter: Xin Zhang (Arizona State University)
We present a kernel-based learning method for predicting DNA-binding proteins from multiple data sources. In cross validation over integrated kernels, the overall average accuracy is over 88%, which is better than previously published results using only structural and sequence information. We further apply the method over proteins with unknown structure information. The results demonstrate that kernel method combining with multiple data sources has great potential in accurately predicting DNA-binding proteins.
Exploiting Advances in ICT Technologies to Affectively Monitor Student Learning
Presenter: Aisha Ijaz (Liverpool Hope University)
This paper examines the potential of exploiting state-of-the art technologies when affectively monitoring student learning to form an enriched and unobtrusive system that responds to humans with the aim of modernizing the way users learn. An amalgamation of affective computing and wireless sensor networks outline an intelligent system integrated within an ambient environment providing a novel model of learning which underpins the interplay between learning and emotions.