Agents that work on a user's behalf at all touch-points - including wireless devices - through which s/he interacts with the web, and provide a range of personalized services that are location, time and device format sensitive. In addition to Intel's grant, we gratefully acknowledge their donation of two workstations, valued at about $8K, to our lab.
Analysis of web content (text and tags), hyperlink structure and usage of web sites to better customize such sites to the needs of site-visitors. E-commerce applications. Relating web analytics to system performance/availability.
Detecting and modeling change/deviations that may indicate computer security/integrity problems.
Finding similarities between XML documents, or between an XML query and a document.
Identifying and extracting useful information from reams of data and thus facilitate smart business decisions. Applications to the computer industry and e-commerce.
Ensemble and hybrid networks; modular approaches for non-stationary problems; multi-classifier systems; combining connectionist and symbolic techniques; applications to difficult object recognition problems. Part of Center for Imaging Science, a joint effort with Washington U., Harvard and MIT.
Analyzing web logs, click-stream analysis, web metrics, recommender systems. Use of knowledge developed in previous tasks to help in a new task. Design of next-generation collaborative filters.
A spatio-temporal problem in which the sequence of symbols emitted by an object may be affected by others in the vicinity.
How can we understand what a neural network does? Rule extraction from neural networks; theory refinement; visualization of network structure and behavior; non-linear methods for dimensionality reduction. Online Demo of Radial Basis Function Network Toolkit
Working in high-dimensions; identifying land cover types from aerial images.
Techniques for identifying outliers and atypical data, and for adapting to dynamic environments. Pose estimation of space objects from image sequences.
Particularly for navigating in hazardous environments and detection environmental hazards.
Dynamic networks are being studied that make occasional judgements on input sequences, and cater to time alignment and dynamic time warping problems. Applied to short-duration underwater signals. A non-linear memory structure based on habituation has been developed that has powerful approximation capabilities.
Modeling of various semiconductor manufacturing processes, and prediction of chip quality.
This study includes practical techniques and theoretical studies on network growth/pruning for valid generalization, study of noise sensitivity of different networks, and performance on limited, high dimensional inputs. Our techniques have been successfully applied to sonar and radar data, and for identification of defects in manufacturing problems.
Adaptive, non-linear regression techniques for forecasting desired response given correlated parameters. Prediction of (chaotic) time series from past samples. Applied to logging data from Schlumberger.
Analysis of records of incoming UT students to predict course loading etc., for improving future course offerings in the College of Engineering.