Abstract. The Self-Organizing Map (SOM) and Learning Vector Quantization (LVQ) algorithms are constructed in this work for variable-length and warped feature sequences. The novelty is to associate an entire feature vector sequence, instead of a single feature vector, as a model with each SOM node. Dynamic time warping is used to obtain time-normalized distances between sequences with different lengths. Starting with random initialization, ordered feature sequence maps then ensue, and Learning Vector Quantization can be used to fine tune the prototype sequences for optimal class separation. The resulting SOM models, the prototype sequences, can then be used for the recognition as well as synthesis of patterns. Good results have been obtained in speaker-independent speech recognition.

Learning-Based Vision and Its Application to Autonomous Indoor Navigation By Shaoyun Chen Adaptation is critical to autonomous navigation of mobile robots. Many adaptive mechanisms have been implemented, ranging from simple color thresholding to complicated learning with artificial neural networks (ANN). The major focus of this thesis lies in machine learning for vision-based navigation. Two well known vision-based navigation systems are ALVINN and ROBIN developed by Carnegie-Mellon University and University of Maryland, respectively. ALVINN uses a two-layer feedforward neural network while ROBIN relies on a radial basis function network (RBFN). Although current ANN-based methods have achieved great success in vision-based navigation, they have two major disadvantages: (1) Local minimum problem: The training of either multilayer perceptron or radial basis function network can get stuck at poor local minimums. (2) The flexibility problem: After the system has been trained in certain r...

Determination of the error rate In the speech database collected here mostly in 1995, there are currently data of 20 speakers and at least four recording sessions of 350 Finnish words for each speaker. The speaker dependent recognition models are trained using three word sets and tested by the remaining one. The error rate given as the result is the number of all phoneme errors (inserted,deleted and changed phonemes) divided by the total number of phonemes. To gain statistical significance for the model comparisons, the tests are normally made for seven different speakers and the error rates are averaged. For verifying the robustness of the models for slightly different speech data also an older database (from 1990) is sometimes used. In general, the older database gives lower average error rates, probably because of the more experienced speakers. For comparisons of the models the post-processing by the Dynamically Expanding Context (DEC) [1]is not applied in order t