Topic: Algorithms for the Estimation of Entropy and Divergence
By: H. Cai (Princeton), S. R. Kulkarni (Princeton), S. Verdu (Princeton)
An important problem in sensor processing is measure the dependence between two or more sensor inputs. This is particularly important and difficult if the sensor geometry and the relationship between the sensors and the environment is unknown. In such cases, clibrating or fusing the sensor data critically dependent on the mutual information between the data streams whose probability distributions are unknown. In this project, we seek universal algorithms (i.e., without assumptions of prior knowledge about the distributions) for estimating key information measures such as entropy, mutual information, and divergence. Preliminary results look very encouraging and should serve as an important tool in the design of algorithms for information in a general Sensorweb environment.
Questions and comments to the webmaster. This page is copyrighted by the Massachusetts Institute of Technology.