🚀 go-pugleaf

Oregon Graduate Institute
Department of Computer Science and Engineering

COLLOQUIUM SERIES


-----------------------------------------------------------------------
Issues in Scientific Computing Colloquium Series
(Co-sponsored by OCATE)

Applications Demand Class-Specific Optimizations: 
       The C++ Compiler Can Do More
Ian G. Angus, Boeing Computer Services

Friday, January 15, 1993,  3:30 p.m.
Main Seminar Room, OGI Administration Building

So far C++ has made few inroads into the realm of scientific computing, 
which is still largely dominated by Fortran.  Of the few attempts that have 
been made to apply C++ to numerically intensive codes, the results have 
often suffered from annoying performance problems.  Examination of these 
problems suggests that they are unlikely to be solved by incremental 
improvements in compiler optimization technology.

In this talk we will:
*	Motivate the discussion by describing a common efficiency problem 
that arises when numerical codes are programmed in C++.
*	Discuss some potential solution strategies that we believe to be 
viable in the near term, but not over the long term.
*	Introduce a minimal mechanism through which a compiler could 
solve these problems by loading class-specific optimizations on an as-
needed basis.  A simple interface that will enable this feature will be 
presented.

While our immediate motivation is that of numerically intensive codes,
our approach will be applicable to all application domains.


-----------------------------------------------------------------------
Getting At Information In Speech
Hynek Hermansky, U S WEST Advanced Technologies

Tuesday, January 19, 1992, 10:30 a.m.
Main Seminar Room, OGI Administration Building

Automatic recognition of speech aspires for automatic decoding of the 
linguistic information from speech.  The task is complicated since, besides 
the linguistic information, speech carries information from many additional 
sources, such as the identity of talker or identity and quality of 
environment in which the speech was produced and over which it was 
transmitted.  The resulting variability in speech is currently being handled 
by extensive training of recognizers on large databases which cover most 
sources of the anticipated variability.

Database collection is expensive, and one can never be sure that all sources 
of variability are being covered.  Our goal is to understand and to be able 
to model (and thus to alleviate) sources of harmful variability in speech.  
We discuss our recent techniques that are aimed at improving 
performance of recognizers in severe telecommunications environments.   
We are primarily guided by engineering considerations.  Nevertheless we 
are arriving to solutions consistent with our knowledge of human 
information processing.


-----------------------------------------------------------------------
The Starburst Database Rule System: Language Design, Implementation
     and Applications
Jennifer Widom, IBM Almaden Research Center

Friday, January 22, 1992, 11:30 a.m.
Room 102, OGI Computer Science and Engineering Bldg.

This talk surveys our development of the Starburst Rule System, a 
production rules facility integrated into the Starburst extensible database 
system at the IBM Almaden Research Center.  Database production rules 
allow definition of operations that are executed automatically whenever 
certain events occur or conditions are met, making the database system 
"active."

The Starburst rule language is based on arbitrary database state 
transitions rather than tuple- or statement-level changes, yielding a clear 
and flexible execution semantics.  The rule system was implemented 
rapidly using the extensibility features of Starburst; it is integrated into all 
aspects of query and transaction processing, including concurrency control, 
authorization, rollback, etc.  Since production rules can be difficult to 
program, we have developed a number of methods for automatically 
generating database rule applications, including integrity constraints, 
materialized views, and deductive rules.


-----------------------------------------------------------------------
Language Issues in Software Specification and Development Colloquium 
Series
(Co-sponsored by OCATE)

Type Inference with Set Constraints
Alex Aiken, IBM Almaden Research Center

Friday, January 22, 1992, 3:30 p.m.
Main Seminar Room, OGI Administration Building

Set constraints are a simple extension of standard type languages.  In 
addition to the usual type constructors, set constraints add the set 
operations (union, intersection, and a kind of negation) and constraints of 
the form "X is a subset of Y".  This combination of inclusion constraints and 
set operations makes it possible to devise some very powerful type 
systems.

While set constraints seem like a natural basis for type systems, the idea 
has not received much attention.  This is partly because many algorithms 
on set constraints are computationally expensive, but also because 
progress has been slow on the central problem of solving systems of set 
constraints.  In this talk, we present an algorithm for solving systems of 
set constraints and discuss its application to type inference.  The talk will 
be illustrated with examples taken from our implementation of a type 
inference system based on solving systems of set constraints. This is joint 
work with Ed Wimmers.