| 1. |
- Isaksson, Alf
(author)
-
Analysis of Identified 2-D Noncausal Models
- 1991
-
Reports (other academic/artistic)abstract
- There are two approaches to the identification of noncausal autoregressive systems in two dimensions differing in the assumed noise model. For both approaches, the maximum likelihood estimator formulated in the frequency domain is presented. The Fisher information matrix is evaluated and found to be the sum of a block-Toeplitz and a block-Hankel matrix. The variance of the parameters, however, cannot be used for comparison of the two approaches, so the variance in the frequency domain is evaluated, assuming that the true system in each case can be described by a model of that type, possibly high-order. In particular, the variance of the spectrum estimate is derived. If the number of parameters tends to infinity, it is shown that the two approaches give the same spectrum estimate variance. The question of which set of true spectra can be described by the respective approaches is discussed.
|
|
| 2. |
|
|
| 3. |
- Isaksson, Alf, et al.
(author)
-
On Recursive Construction of Trees as Models of Dynamical Systems
- 1991
-
In: Proceedings of the 30th IEEE Conference on Decision and Control. - Linköping : Linköping University. - 0780304500 ; , s. 1686-1687 vol.2
-
Conference paper (peer-reviewed)abstract
- An issue that is of importance for control applications is discussed: how to construct the trees online, i.e. recursively, as more and more data become available. A theorem regarding recursive tree-building is stated and proved, and implementation issues are considered.
|
|
| 4. |
- Isaksson, Alf
(author)
-
System Identification Subject to Missing Data
- 1991
-
In: Proceedings of the 1991 American Control Conference. - Linköping : Linköping University. - 0879425652 ; , s. 693-698
-
Conference paper (peer-reviewed)abstract
- In this paper we study parameter estimation when the measurement information may be incomplete. As a basic system representation we use an ARX-model. The presentation covers both missing output and input. First reconstruction of the missing values is discussed. The reconstruction is based on a state-space formulation of the system, and is performed using the Kalman filtering or fixed-interval smoothing formulas. Several approaches to the identification problem are then presented, including a new method based on the so called EM algorithm. The different approaches are tested and compared using Monte-Carlo simulations. The choice of method is always a trade off between estimation accuracy and computational complexity. According to the simulations the gain in accuracy using the EM method can be considerable if much data are missing.
|
|
| 5. |
- Strömberg, Jan-Erik, et al.
(author)
-
Neural Trees : Using Neural Nets in a Tree Classifier Structure
- 1991
-
In: Proceedings of the 1991 International Conference on Acoustics, Speech, and Signal Processing. - Linköping : IEEE. - 0780300033 ; , s. 137-140
-
Conference paper (peer-reviewed)abstract
- The concept of tree classifiers is combined with the popular neural net structure. Instead of having one large neural net to capture all the regions in the feature space, the authors suggest the compromise of using small single-output nets at each tree node. This hybrid classifier is referred to as a neural tree. The performance of this classifier is evaluated on real data from a problem in speech recognition. When verified on this particular problem, it turns out that the classifier concept drastically reduces the computational complexity compared with conventional multilevel neural nets. It is also noted that these data make it possible to grow trees online from a continuous data stream.
|
|
