Past Research Interests and Activities (1979-2000)
Learning Problem-solving Heuristics
My thesis Research (1976-1979, See IJCAI 1979)
The Massively Parallel Dado Machine(~1979-~1989)
Current status: DADO has been licensed to Fifth Generation Computer
Corporation (and abruptly left academia for its obvious commercialization.) The
final chapter on DADO has not yet been written! Stay tuned!
Deductive and Expert Databases
ACE - The First Deductive (Expert)
Database System (1980 - ~1982)
Active/Deductive Database System Reorganization and
"Predictive" Load Balancing (1990 - ~1994)
The PARADISER system (1990-~1994)
Parallel Rule-based Systems (Then and Now) (1980 - ~1994)
Incremental Rule Processing in Parallel Environments
(1990 - ~1994)
"Real-world" Application Studies
ALEXSYS - Mortgage-backed Security
Trading in Parallel. (1988 - ~1990)
Speeding Up Monte Carlo Simulations (~1993)
I've also studied parallelization strategies for Monte Carlo
simulations. Check out Dynamic Neighborhood Bounding, a simple heuristic
strategy to save on expensive function computations that is also amenable to
fast parallel execution. (Follow the link to "papers".)
MERGE-PURGE: Intelligently Integrating Heterogenous Databases
(~1993-1995)
Although the term merge/purge is used by our Citicorp
collaborators and other commercial organizations, it is probably better termed The Data Scrubbing, or Data Cleaning Problem.
(That's a term Mike Stonebraker used to describe this problem--and
Stonebraker-terminology is almost always colorful and illuminating.)
Our success on data supplied
by the Child Welfare Department of the State of Washington is memoralized in
the letter viewable by clicking here.
Integrating Multiple Models: AAAI96 Workshop
Many researchers are working on techniques for integrating
multiple learned models. So Phil Chan, David Wolpert and myself organized a
Workshop at AAAI-96 called Integrating
Multiple Models for Improving and Scaling Machine Learning.
My small part
in US DOJ Versus Microsoft
White Paper to NSF Workshop on R&D Opportunities in Federal IS
JAM: Java Agents for Meta-learning
We've recently won a number of grants for applying our
research on Data Mining and Meta-learning to the problem of fraud and intrusion
detection in financial information systems. This new project involves building
a Java-based agent infrastructure for learning over distributed databases.
Check out The JAM PROJECT
page for details.
META-LEARNING for Scalable Data Mining
Meta-learning is a term we
coined for an approach gaining in popularity in the DM/KDD community these
days. The essential idea is to combine a number of separately learned
classifiers or models in such a fashion that machine learning can be scaled to
large (and inherently distributed) databases, and (ideally) accuracy can be
boosted. But how does one combine separately learned classifiers? We consider
that question as a learning problem and apply machine learning algorithms to
"meta-learn" how underlying classifiers behave or correlate with
eachother. Hmmmmm...
ELECTRONIC COMMERCE and FINANCIAL INFORMATION SYSTEMS: Widely
Distributed Data Mining and Fraud Detection
My current interest lies in the future of large scale
network-based information systems, especially Fraud and
Intrusion Detection. We are studying methods to apply learning agents, and
Meta-learning agents in WAN financial information systems to learn how to
detect fraudlent transactions. This work is directly relevant to Information
Warfare problems that will undoubtedly occur in the future. (These problems
probably occur today but no one has yet learned how to detect it globally! Stay
tuned!)
This research is being conducted in collaboration with our friends at the Financial Services Technology Consortium.,
KNOWLEDGE DISCOVERY in DATABASE and DATA MINING (KDD/DM)
My former days of expert systems research, and parallel
hardware design and implementation were great fun. But now I am specializing in
parallel/distributed processing for Machine Learning
(or alternatively
click here), Intelligent
Information Systems, Knowledge
Discovery in Databases and Data Mining with particular application to
Financial Information Systems, and Intrusion Detection Systems. Check out
recent events like KDD2000,
that I co-chaired.
Collaborators
Financial Services
Technology Consortium (FSTC)
I've been working with the Financial Services Technology
Consortium (FSTC) having played some small role in its early formation. The
FSTC's "Fraud Project" is the focus of our joint effort. See
the description on Fraud
and Intrusion Detection in Financial Information Systems using Meta-learning
Agents for details about our approach.
My relationship with Citicorp was established through my
consultancies and research relationships with the Citicorp Technology Office.
Working with companies like Citicorp is a wonderful opportunity to learn about
real-world problems and datasets for use in establishing benchmarks and
controlled studies of fundamental technologies.
Chase has been the primary mover behind the fraud project
and has supplied us with valuable information and data to conduct our work.
First Union has been a valuable collaborator and also has
supplied us with valuable information and data.
My collaboration with Citicorp and other organizations lead to the development of a number of fundamental technologies that were demonstrated in "real-life" financial applications. Some of these are described in the "prior research" web page. Seek them out.
iPrivacy – Private Ecommerce
on the Web (1996-2001)
As few know, a few friends and co-founders developed an internet privacy company way ahead of it’s time called iPrivacy, not to be confused with a more recent company with that name dealing with privacy products to secure personal information. Our iPrivacy developed a complete private surfing, shopping and shipping system that was ready for test in 2001. We had a contract with the US Post Office, a large commercial bank who signed up over 30,000 customers in Silicon Valley, and a fully fielded system ready for use. 911 had made other plans for our company which soon shut down in the wake of the nation-wide crisis.
What remains of iPrivacy is a body of excellent work not published in the academic literature, but remaining on record in the US Patent and Trademark Office:
Patents Issued:
|
PAT. NO. |
Title |
||
|
1 |
|||
|
4 |
|
Method and system for obscuring user access patterns using a buffer memory |
|
|
7 |
|
||
Applications still in prosecution:
|
PUB. APP. NO. |
Title |
|
|
1 |
||
|
2 |
Method and system for private shipping to anonymous users of a computer network |
|
|
3 |
||
|
4 |
Method and system for private shipping to anonymous users of a computer network |
Better have a postscript viewer configured before
downloading!
Sponsors:
New York State Science and Technology Foundation CAT Program
