 |
Independent Component Analysis
Independent Component Analysis (ICA) is a well-known problem in the fields of neural
networks and signal processing. It is closely related to blind sources separation. The
purpose of ICA is to extract statistically independent components from a set of dependent
data.
|
|
Time Series Prediction
Neural networks can be used to build non-linear predictors which perform better tha linear
standard prediction tools in most situations. Extracting adequate auto-regressive vectors
from a series is a key problem when dealing with non-linear predictors.
|
|
Smart Sensors
Smart sensors are electronic devices combining the sensors themselves and some electronic
components (pre-)processing their outputs. The advantages of smart sensors reside in
decreased noise and interferences, but also in portability and production costs.
Independent component analysis can be used in smart sensors.
|
|
Curvilinear Distance Analysis
Principal Component Analysis (PCA) is a frequently used technique in the field of data analysis.
PCA allow to find important
directions in a set of numerical data. However, PCA is a linear method which
fails when the
data set follows a curved hypersurface instead of an hyperplane. Curvilinear Component Analysis
(CCA) and Curvilinear Distance Analysis (CDA) are then used to project the data
set: both methods allow nonlinear projection of curved structures.
|
|
VLSI implementation of Neural Networks and Fuzzy Systems
Some neural networks and fuzzy logic algorithms benefit from a dedicated ASIC
(Application-Specific Integrated Circuit) implementation. Advantages can be found not only
in the performances (number of operations per second,...), but also (and mainly) in the
portability of the systems, the low power consumption, no necessity of cumbersome computer
or signal processing device, etc. Analog ASICs are particularly adapted to the implementation of
neural networks and fuzzy systems.
|
