Dynamic Graph Analysis and Optimal Control -
Dynamic Programming Approaches to Safety
Ashira Khera, San Jose State University.
The use of dynamic programming (DP) for finding
optimal paths in graphs is a widely studied and well understood
problem. Several practical applications of this technique include
Internet routing, convolutional error-correcting codes and trust
management. In this research project, considering a dynamical
graph structure, we focus on optimal control problems over
it and solve them with DP. In particular, with regards to the
objective function used in the DP scheme, we study how the
solution to the problem changes when we use a multiplicative or
an additive function as the optimization criterion. Starting with a
controlled, continuous dynamical model, by properly discretizing
the dynamics and the control space we can approximate it with
a system evolving on a control-dependent graphical structure.
We will then consider this graphical structure over a finite time
horizon to find solutions for certain optimal control problems,
(for instance, min-time problems) and compare the performance
of the two aforementioned optimality criteria. We shall finally
look into a safety problem, interpret it as an optimization one,
and relate it in an original way to some known security issues.
Feature Extraction and Analysis for Body Sensor Networks
Amalia Viti, Columbia University.
Wireless body sensor networks are networks composed of nodes placed on different parts of the human body. These nodes are able to communicate in a wireless manner and transmit data from sensors mounted on them to a base unit or a computer that analyzes the data received. Wireless body sensor networks make on-body and mobile health-care monitoring possible; such systems can integrate information from different sources, and can initiate actions or trigger alarms when needed.
In this project we investigate a collaborative signal processing scheme for physical movement monitoring with motion sensors. The signal processing consists of preprocessing, feature extraction and classification. We define a measure on feature significance as well as features’ correlations. Because these nodes have a limited amount of battery power, we extract features to characterize the data. We develop a metric to rank each feature by its significance with respect to a particular movement class. We validate our metric by analyzing the accuracy of classification when using a subset of features selected by our metric. Using the characteristics of the best features we find from the rankings, we develop new features to improve the system accuracy and then assess their effectiveness.
|
