|
1 - Arzucan Ozgur , Dragomir Radev
Semi-Supervised Classification for Extracting Protein Interaction Sentences
Abstract:
We introduce a relation extraction method to identify the sentences in biomedical text that indicate an interaction among the protein names mentioned. Our approach is based on the analysis of the paths between two protein names in the dependency parse trees of the sentences. Given two dependency trees, we define two separate similarity functions (kernels) based on cosine similarity and edit distance among the paths between the protein names. Using these similarity functions, we investigate the performances of two classes of learning algorithms, Support Vector Machines and k-nearest-neighbor, and the semi-supervised counterparts of these algorithms, transductive SVMs and harmonic functions, respectively. Significant improvement over the previous results in the literature is reported as well as a new benchmark dataset is introduced. Semi-supervised algorithms perform better than their supervised version by a wide margin especially when the amount of labeled data is limited.
|
|
2 - Kyla McMullen , Elliot Soloway
Fostering Concept Learning in Low Socio-Economic Environments Through The Use of Educational Software
Abstract:
Research has shown that the main difference between children of lower and higher socio-economic status is that the children in lower socio-economic environments have poor categorization skills. Categorization skills are essential in concept learning and language proficiency. There are also a wide range of educational software and gaming features that are useful in facilitating an entertaining game environment. This project's goal is to embed the mechanisms involved in categorization and concept learning within a game environment, thus providing children with a fun and effective system of attaining categorization skills.
|
|
3 - Julie Weber , Martha E. Pollock
This proposal describes my work in the development of an intelligent cognitive
Abstract:
This proposal describes my work in the development of an intelligent cognitive orthotic that interacts with its user in a personalized fashion, deciding whether, when and how to interact with its user based on that user's needs and preferences. I consider two types of users: those who work in an office environment and require assistance with managing their daily meeting and project schedule, and users with cognitive disabilities who require guidance in performing their daily tasks; and I examine various interaction types including reminders, requests for confirmation of a task having been completed, requests for permission to assist in the performance of a task, and requests for feedback. The system I develop will have the ability to learn user preferences from both explicit and implicit user feedback in the process of adapting its interactions to meet the needs of each individual user
|
|
4 - Erik Talvitie , Satinder Singh Baveja
Building Incomplete But Accurate Models
Abstract:
An agent attempting to model a sufficiently complex environment may wish to reduce the amount of modeling effort necessary by restricting the questions it attempts to answer about the world. In particular, we imagine an agent that seeks to answer questions on an abstract level by not distinguishing between certain observations. In order to make accurate predictions regarding the aggregate observations of interest however, the agent may still need to model the world at a finer resolution. We provide worst-case bounds on the difficulty of taking a partition of the observation space of the world and finding a refinement of that partition such that a model of the refined aggregate system will make accurate predictions for the original questions of interest. Such a model may be significantly more compact than a complete model of the world. Our results apply generally to all discrete, finite-dimensional dynamical systems.
|
|
5 - Smita Krishnaswamy , Igor Markov, John Hayes
AnSER: A Fast Reliability Evaluator for Logic Circuits
Abstract:
We are developing an automated computer-aided diagnosis (CAD) system for classification of lung nodules as malignant or benign on CT scans. The CAD system involves segmentation, feature extraction, and classification. Segmentation was performed with a 3D active contour (3DAC) method, and malignant and benign nodules were differentiated by their morphological and texture characteristics. The performance of the computer classifier was comparable to that from radiologists' assessments of the likelihood of malignancy (two-tailed p = 0.98).
|
|
6 - Ted Way , Heang-Ping Chan , Jeffrey A. Fessler
Computer-aided diagnosis of lung nodules
Abstract:
We are developing an automated computer-aided diagnosis (CAD) system for classification of lung nodules as malignant or benign on CT scans. The CAD system involves segmentation, feature extraction, and classification. Segmentation was performed with a 3D active contour (3DAC) method, and malignant and benign nodules were differentiated by their morphological and texture characteristics. The performance of the computer classifier was comparable to that from radiologists' assessments of the likelihood of malignancy (two-tailed p = 0.98).
|
|
7- Hoeryong Jung , Doo Yong Lee
Surgery Planning Simulator for the Closed Reduction and Internal Fixation
Abstract:
This presentation presents a surgery planning simulator for the closed reduction and internal fixation surgery. The developed simulator employs 3D femur model which is reconstructed from actual patient's CT data and 3D screw models generated using solid modeling software (Pro-e). The simulator also employs 2D mouse as an input device to interact with the femur model simply and intuitively. Centerline of the femur neck is extracted to assist the user to determine accurately an insertion point and direction of the screws. The user can confirm position of the screws at any cross section she/he selects. Also, the simulator can show the post-operative appearance of the femur. The simulator allows the user to plan both the multiple pinning (MP) and dynamic hip screw (DHS) procedures. The simulator allows the surgeon to determine the insertion point and direction of the screw precisely on the femur model that was impossible in conventional planning method.
|
|
8- Arvind Rao , Alfred O. Hero III, David J. States , James Douglas Engel
Understanding Long Range Transcriptional Regulation via Integrative Genomics
Abstract:
How will clinical epidemiology look in 2017? Recent events have sparked global concern over the current state of infectious disease control. The accurate prediction of health and disease in a cohort of subjects that are loosely monitored for changes in individual biomarkers, disease spread through social interactions, and epidemiological patterns is an elusive goal in clinical medicine. We introduce a predictive Bayesian framework for in-situ predictive health and disease that can attain this goal. The analysis engine consists of several components: a graph-theoretic social engine that takes into consideration knowledge of contacts of the patient and community structure of the social network, a diagnosisengine based on ensemble of tree classifiers that considers the pathogen exposure and the molecular response parameters, and a dynamic Bayesian network prediction engine that integrates past and present social and diagnostic information for accurate prediction of a individual's health.
|
|
9- Matthew Reyes , David Neuhoff
Compression of a Binary Image Using a Markov Random Field Model
Abstract:
Markov random fields (MRFs) are used extensively in many areas including image processing, pattern recognition and sensor networks. Their popularity stems from their equivalence with Gibbs random fields, and the consequent tractability and structure-imposing nature of the distribution. In image processing, MRFs have been used primarily in segmentation, texture analysis and inference. Here we discuss the use of Markov random fields in a bilevel image compression scheme. We losslessly transmit a subset of the image pixels and use two different MRF models to reconstruct the missing (non-transmitted) image pixels. In addition to the theoretical solution provided by the form of the MRF distribution, we demonstrate the practicability of our method by applying it to some binary images. We discuss how the method presented can be extended to the compression of gray-scale images and how message passing can be incorporated int
|
|
10 - Byoungho Kwon , Young Jin Park
Sound Source Localization using the Compensation Method in Robot Platform
Abstract:
While various methods for sound source localization have been developed, most of them utilize on the time difference of arrival (TDOA) between microphones or the measured head related transfer functions (HRTF). In case of a real robot implementation, the former has a merit of light computation load to estimate the sound direction but can not consider the effect of platform on TDOAs, while the latter can, because characteristics of robot platform are included in HRTF. However, the latter needs large resources for the HRTF database of a specific robot platform. We propose the compensation method which has the light computation load while the effect of platform on TDOA can be taken into account. The proposed method is used with spherical head related transfer function (SHRTF) on the assumption that robot platform, for example a robot head, installed microphones can be modeled to a sphere. We verify that the proposed method decreases the estimation error caused by the robot platform throug
|
