Computer Science

The research described here is part of a larger project with the
objective of determining if a writer believes a proposition to
be true or false. This task requires a deep understanding of
a proposition’s semantic context, which is far beyond NLP’s
state of the art. In light of this difficulty, this paper presents a
shallow semantic framework that addresses the sub-problem
of finding a proposition’s truth-value at the sentence level.
The framework consists of several classes of linguistic elements
that, when linked to a proposition through specific
lexico-syntactic connectors, change its truth-value. A pilot
evaluation of a system implementing this framework yields
promising results.

User generated content is extremely valuable for mining
market intelligence because it is unsolicited. We study
the problem of analyzing users’ sentiment and opinion
in their blog, message board, etc. posts with respect
to topics expressed as a search query. In the scenario
we consider the matches of the search query terms are
expanded through coreference and meronymy to produce
a set of mentions. The mentions are contextually
evaluated for sentiment and their scores are aggregated
(using a data structure we introduce call the sentiment
propagation graph) to produce an aggregate score for
the input entity. An extremely crucial part in the contextual
evaluation of individual mentions is finding which
sentiment expressions are semantically related to (target)
which mentions — this is the focus of our paper.
We present an approach where potential target mentions
for a sentiment expression are ranked using supervised
machine learning (Support Vector Machines) where the
main features are the syntactic configurations (typed dependency
paths) connecting the sentiment expression
and the mention. We have created a large English corpus
of product discussions blogs annotated with semantic
types of mentions, coreference, meronymy and sentiment
targets. The corpus proves that coreference and
meronymy are not marginal phenomena but are really
central to determining the overall sentiment for the toplevel
entity. We evaluate a number of techniques for
sentiment targeting and present results which we believe
push the current state-of-the-art.

This paper presents a rich annotation scheme for mentions, co-reference, meronymy, sentiment expressions, modifiers of sentiment expressions including neutralizers, negators, and intensifiers, and describes a large corpus annotated with this scheme.  We describe how this corpus relates to recent, state-of-the-art work in sentiment analysis, and define the various annotation types, provide examples, and show statistics on occurrence and inter-annotator agreement.  This resource is the largest sentiment-topical corpus to date and is publicly available.  It helps quantify sentiment phenomena, and allows for the construction of advanced sentiment systems and enables direct comparison of different algorithms.

As smartphones become more pervasive, they are increasingly targeted by malware. At the
same time, each new generation of smartphone features increasingly powerful onboard sensor
suites. A new strain of ‘sensor malware’ has been developing that leverages these sensors to steal
information from the physical environment — e.g., researchers have recently demonstrated how
malware can ‘listen’ for spoken credit card numbers through the microphone, or ‘feel’ keystroke
vibrations using the accelerometer. Yet the possibilities of what malware can ‘see’ through a
camera have been understudied.
This paper introduces a novel ‘visual malware’ called PlaceRaider, which allows remote at-
tackers to engage in remote reconnaissance and what we call “virtual theft.” Through completely
opportunistic use of the phone’s camera and other sensors, PlaceRaider constructs rich, three
dimensional models of indoor environments. Remote burglars can thus ‘download’ the physical
space, study the environment carefully, and steal virtual objects from the environment (such
as financial documents, information on computer monitors, and personally identifiable informa-
tion). Through two human subject studies we demonstrate the effectiveness of using mobile
devices as powerful urveillance and virtual theft platforms, and we suggest several possible
defenses against visual malware.

Distributed Hash Tables (DHTs) such as Chord and Kademlia offer an efficient solution for locat-
ing resources in peer-to-peer networks. Unfortunately, malicious nodes along a lookup path can easily
subvert such queries. Several systems, including Halo (based on Chord) and Kad (based on Kademlia),
mitigate such attacks by using a combination of redundancy and diversity in the paths taken by redun-
dant lookup queries. Much greater assurance can be provided, however. We describe Reputation for
Directory Services (ReDS), a framework for enhancing lookups in redundant DHTs by tracking how
well other nodes service lookup requests. We describe how the ReDS technique can be applied to vir-
tually any redundant DHT including Halo and Kad. We also study the collaborative identification and
removal of bad lookup paths in a way that does not rely on the sharing of reputation scores — we show
that such sharing is vulnerable to attacks that make it unsuitable for most applications of ReDS. Through
extensive simulations we demonstrate that ReDS improves lookup success rates for Halo and Kad by
80% or more over a wide range of conditions, even against strategic attackers attempting to game their
reputation scores and in the presence of node churn.

In this assignment, a simple system for image warping and stereo matching techniques is implemented. For the warping functions, the implementation provides rotation and projective transformation by using bilinear interpolation. For stereo matching, the implementation provides a simple block-based matching technique, a HMM technique using dynamic programming, a belief propagation technique using min convolution, and a graph cuts approach based on ��-�� swap move.

Abstract Attributes are visual concepts that can be detected by machines, understood by humans, and shared across categories. They are particularly useful for fine-grained domains where categories are closely related to one other (eg bird species recognition). In such scenarios, relevant attributes are often local (eg ���white belly���), but the question of how to choose these local attributes remains largely unexplored.

Abstract We introduce a new approach for part-based human pose estimation using multi-layer composite models, in which each layer is a tree-structured pictorial structure that models pose at a different scale and with a different graphical structure. At the highest level, the submodel acts as a person detector, while at the lowest level, the body is decomposed into a collection of many local parts. Edges between adjacent layers of the composite model encode cross-model constraints.

Despite the proliferation of mobile health applications, few target low-literacy users. This is a matter of concern because 43% of the United States population is functionally illiterate. To empower everyone to be a full participant in the evolving health system and prevent further disparities, we must understand the design needs of low-literacy populations. In this paper, we present two complementary studies of four graphical user interface (GUI) widgets and three different cross-page navigation styles in mobile applications with a low-literacy, chronically-ill population. Participant's navigation and interaction styles were documented while they performed search tasks using high fidelity prototypes running on a mobile device. Results indicate that participants could use any non-text based GUI widgets but preferred and performed efficiently with large radio buttons. For navigation structures, users performed best when navigating a linear structure, but preferred the features of cross-linked navigation. Based on these findings, we provide some recommendations for designing accessible mobile applications for low-literacy populations.

Being aware of one's portion sizes is a key component of maintaining a healthy diet, however, it is difficult for individuals especially low literacy populations to estimate their consumption. Nutritional monitoring applications can help but most of them are designed for people with high literacy and numeracy skills. In this paper, we designed and evaluated six portion size estimation interfaces through a Wizard of Oz based experiment using low-fidelity prototypes with ten varying literacy individuals. The interfaces were designed based on the cognitive strategies adults use for reporting portion sizes in diet recall studies. Participants made correct estimates with interfaces designed for liquid and amorphous foods, but had difficulties with those designed for solid foods. Based on these findings, we provide recommendations for designing accurate and low literacy-accessible portion size estimating mobile interfaces.

Documents retrieved in response to a user's query should reflect the intention of the user. Keyword searches are not sufficient to accomplish this task. This paper considers Latent Semantic Indexing (LSI) as possible solution. LSI is considered in the context of a large set of World Bank documents. LSI is shown to be useful in this regard, but not a complete solution. Additional, domain specific, information would be needed as well.

This paper includes several reasons for the underrepresentation of women in computing, and then describes two low-cost project instances that address the reasons for the decline in women's enrollment in computing classes. One project spans seven semesters from fall 2000 to spring 2006 at a small liberal arts school; the other, the spring 2006 semester at a large research institution. Concluding sections provide anecdotal and statistical evidence that the project is successful over two time periods within two diverse schools.

Starting with the 2003 academic year, the advanced placement (AP) exams in computer science conducted by the college board moved from a C++ to a Java based curriculum. In order to assist high school computer science teachers with the transition from C++ to Java, the ACM together with the college board conceived of Java Engagement for Teacher Training (JETT) workshops. In the Fall of 2003, Indiana University organized such a workshop. The outreach workshop was held over two days and was attended by 35 high school computer science teachers from Indiana and several nearby states. The workshop was conducted as an inter-disciplinary effort with faculty and support from Computer Science, Information Systems, Informatics, and the School of Library and Information Science. Sessions spanned the gamut of foundational concepts of object-oriented programming in Java to network games. A theme of the workshop was to address the problem of the ever-shrinking pipeline of women in the IT field. With this in mind, we also conducted a separate session titled "Where have all the women gone?" Judging by participant feedback, the workshop was a resounding success. This paper describes our experience in organizing the JETT workshop, the lessons learned, and outlines our plans for the future to build upon this K-16 relationship-building exercise.

We present performance results on a Windows cluster with up to 768 cores using MPI and two variants of threading - CCR and TPL. CCR (Concurrency and Coordination Runtime) presents a message based interface while TPL (Task Parallel Library) allows for loops to be automatically parallelized. MPI is used between the cluster nodes (up to 32) and either threading or MPI for parallelism on the 24 cores of each node. We use a simple matrix multiplication kernel as well as a significant bioinformatics gene clustering application. We find that the two threading models offer similar performance with MPI outperforming both at low levels of parallelism but threading much better when the grain size (problem size per process) is small. We find better performance on Intel compared to AMD on comparable 24 core systems. We develop simple models for the performance of the clustering code.