CGL Meeting Agenda

Wednesday, August 2, 1995

Location:: DC 1304
Time:: 1:30 PM
Chair:: Wolfgang Heidrich

1. Adoption of the agenda - additions/deletions thereto.

2. Coffee Hour and Next Meeting:

Coffee hour this week:

???

Coffee hour next week:

???

Next Meeting

Date:: August 9, 1995
Time:: 1:30 PM
Location:: DC 1304
Chair:: Fabrice Jaubert
Tech Presentation:: Nicolas Durand

Forthcoming: (list next 4 and trades)

Chairs:: Rick Kazman; Nathan Konrad; Rob Kroeger; Wayne Liu
Tech presentations:: Saar Friedman; Wolfgang Heidrich; Fabrice Jaubert; Rick Kazman

3. Technical Presentation:

Presenter:: Don Dragomatz

Title:: NC Toolpath Generation using Pixels Models and Space-filling Curves

Abstract:: A method for creating paths for computer-controlled machining is described that is based on Hilbert space-filling curves and pixel images of the object to be machined. The method is compared to traditional roughing and finishing path generation methods as well as pixel-only schemes. The method shows promise in terms of computational effort and support for general path generation (roughing and finishing). However, it does not handle arbitrary exclusion zones well and does not produce paths with pleasing topologies. Some discussion of possible solutions to these problems will therefore be given as well.

4. General Discussion Items:

Richard introduces Thomas Pflaum as a new CGL member

5. Action List (remember to update AL_active)

U. of T. visit (ask Rob)

6. Directors Meeting:

7. Seminar(s):

The University of Waterloo 200 University Avenue Waterloo, Ontario The Institute for Computer Research (ICR) Presents a Colloquium on A Multi-Speaker Putonghua (Spoken Chinese) Recognizer by: Dr. Chorkin Chan of: Department of Computer Science Hong Kong University Hong Kong Island Date: Wednesday, August 9, 1995 Time: 3:30 pm. Place: William G. Davis Computer Research Centre, Room 1302 Abstract: This talk reports the work being done at the University of Hong Kong on a system consisting of 20 speaker-dependent continuous Putonghua recognizers. The idea of a multi-speaker system is to select the most appropriate speaker-dependent recognizer to serve a particular user. Such speaker-dependent recognizers will be adapted to improve its compatibility. Left/right triphones are concatenated to form triphones which are nodes of an HMMnet. Each semi-triphone is a tied mixture HMM. Recognizing an unknown utterance is to perform a beam search over the network for speed and using the two-pass A* search to yield the n-best candidates. These candidates will be post-processed by a language model of bigram statistics of word semantic classes. Everyone is welcome. Refreshments served.