|
Summary
Applied machine-learning researcher.
Stanford PhD with 10 years experience at major Internet portal
and leading industrial research lab.
Passion for working on challenging real-world problems.
Track record of innovative solutions.
Extensive system-building experience.
Application areas include natural language processing, information
retrieval, and data mining.
|
Employment
Terra Lycos, Waltham, MA
Research Scientist, 1999–2002
Projects used in live service on Lycos site:
|
- Product-level targeted ads:
Given a user's search query, present a customized ad
recommending individual products or groups of products
from an advertiser's inventory.
Increased referrals to Barnes & Noble.com
by over 300%.
3,000 lines of Gawk and C-shell scripts.
| |
Uses domain knowledge to match query to inventory
"intelligently". To meet run-time requirements,
anticipates all queries that might plausibly be
asked about the inventory,
and precomputes best responses off-line.
|
- Classification into a very large (400,000-node)
hierarchy:
Given a user's search query, find the topics in the hierarchy
that are most relevant, as well as the most relevant documents
stored under each topic.
Increased clickthroughs on topics by 20%.
10,000 lines of C.
| |
Applies novel taxonomic reasoning rules to generalize
and prune the set of candidate topics returned,
and to overcome spam and other artifacts.
|
- Tours of web sites:
Present an automatically-advancing sequence of web sites
to users.
Spawned two high-page-view products:
Lycos Celebrity Slide Shows, and Matchmaker Member
Photo Previews.
3,000 lines of HTML and client-side JavaScript.
Two-person team.
| |
Includes prefetching of web sites, automatic tour pausing
if the user clicks on a site in the tour, and
a DHTML-based drag-and-drop editor that allows users
to create their own tours without downloads.
|
Infrastructure projects:
|
- Search engine evaluation:
Developed:
(1) Automated method that detects variations
in search-result quality over time;
(2) System that lets humans do blind evaluation
of search results.
Tests became part of Lycos's standard "Service Level Agreement"
contract with search providers.
Automated method is 2,000 lines of Gawk and C-shell scripts.
Blind evaluation is 3,000 lines of Jscript.
| |
Automated method leverages human judgements of search-result
quality to the extent that they are already available
(in indirect form) on the web.
|
- Author classifier:
Determine whether a user's query is for a book author.
Used for generating a weekly report for
Barnes & Noble.com
on trends in author popularity.
3,000 lines of Perl, Gawk, and C-Shell scripts.
| |
Compiles an extensive author list
through information extraction from the web.
Includes name-matching heuristics, spelling correction,
and filtering of "quasi-authors".
|
- Alteration Server:
Infrastructure for creating and serving altered versions
of web pages on-the-fly.
Enables controlled experiments to be run in live service,
thus allowing page design and content
to be continually improved.
200 lines of Jscript, 2,000 lines of C,
and 1,500 lines of Mod Perl.
Two-person team.
| |
Modular architecture works for any server environment.
Typically adds just 80-100 millisecs to serve a variant of a page.
|
- User profile cookie:
Lightweight algorithm for building up
a profile of each user’s interests,
thus enabling personalization throughout the site.
400 lines of C.
| |
Approximates recency/frequency counts for the user's visits
to different parts of the site so as to minimize cookie size
while maintaining a transparent representation.
|
Other projects and responsibilities:
|
- Data mining:
Numerous data-mining projects, including analysis of
user habituation, association of keywords with web pages,
and demonstration that a particular ad deal
was eroding the Lycos user base.
This last demonstration was
the basis for Lycos’s discontinuation of the deal.
|
- Lycos Innovation Committee:
Chairman of Committee.
Responsible for collecting, evaluating, and processing
patent disclosures from Lycos employees.
|
- Technical due diligence:
Evaluated technology of a few dozen companies for
potential partnerships or acquisitions.
|
Mitsubishi Electric Research Laboratories, Cambridge, MA
Research Scientist, 1992–1999
Projects include:
|
- Indoor navigation:
Handheld device equipped with a variety of sensors.
Navigates indoors by learning distinctive patterns
of sensor readings exhibited by staircases, elevators,
steel beams, etc.
15,000 lines of Java.
Two-person team.
| |
Maintains probabilistic model of user's location
at all times, with Bayesian updates based on latest
sensor readings.
|
- Context-sensitive spelling correction:
The task of fixing spelling mistakes
that happen to result in a valid word —
a peace of cake, not to difficult, etc.
40,000 lines of C and C-shell scripts.
Interoperable across Linux, HPUX, Irix.
| |
Formalized problem as word prediction in
high-dimensional space (10,000+ features).
Using an architecture based on Winnow algorithm,
achieved highest accuracy in literature.
|
- English Writer’s Assistant:
Grammar checker designed to fix characteristic errors
made by Japanese speakers writing in English.
24,000 lines of C, 3,000 lines of Tcl/Tk.
Three-person team.
| |
Includes tagger,
noun-phrase parser,
grammar correction modules (e.g., for missing determiners),
and language tools (e.g., context-sensitive dictionary lookup).
|
- Social virtual reality:
Virtual environment to facilitate interaction among people.
Integrates gesture recognition, speech recognition,
natural language understanding, rule-based reasoning,
speech generation, and physics-based animation.
Eight-person team.
| |
Designed for applications such as learning a foreign language
by "virtual immersion" in the culture.
|
Bell Laboratories, Murray Hill, NJ
Member of Technical Staff, Summer 1987
Preliminary work on name pronunciation with Ken Church
and Mark Liberman.
|
Education
Stanford University, Stanford, CA
PhD in Computer Science, 1991
| Thesis: |
Pronouncing names by a combination of
rule-based and case-based reasoning |
Edinburgh University, Edinburgh, Scotland
MPhil in Artificial Intelligence, 1984
| Thesis: |
Lexicrunch: An expert system
for word morphology |
Princeton University, Princeton, NJ
BSE in Electrical Engineering and Computer Science, 1982
|
Awards
Best Paper Award at AAAI
Best Paper Award from JAVIOS
AT&T Bell Laboratories PhD Fellowship
Marshall Scholarship
NSF Fellowship
Valedictorian at Princeton University
|
Publications
|
Internet
- Andrew R. Golding.
Returning databases as search results.
2001.
Unpublished.
|
Context-sensitive spelling correction
- Andrew R. Golding and Dan Roth.
A Winnow-based approach to
context-sensitive spelling correction.
Machine Learning 34 (Feb 1999), 107–130.
Special issue on Natural Language Learning.
|
- Andrew R. Golding and Dan Roth.
Applying Winnow to context-sensitive spelling correction.
In Machine Learning: Proc. 13th
International Conference, Bari, Italy, 1996.
Pages 182–190.
| |
One of 12 papers selected for plenary session.
|
- Andrew R. Golding and Yves Schabes.
Combining trigram-based and feature-based methods
for context-sensitive spelling correction.
In Proc. 34th Annual Meeting
of the Association for Computational Linguistics,
Santa Cruz, CA, 1996.
Pages 71–78.
|
- Andrew R. Golding.
A Bayesian hybrid method for context-sensitive
spelling correction.
In Proc. Third Workshop on Very Large Corpora,
Cambridge, MA, 1995.
Pages 39–53.
|
Case-based reasoning and name pronunciation
|
|
My PhD was on Anapron,
a system for pronouncing surnames, based on
a general architecture for combining rule-based
and case-based reasoning.
|
- Andrew R. Golding and Paul S. Rosenbloom.
Improving accuracy by combining rule-based
and case-based reasoning.
Artificial Intelligence 87 (Nov 1996),
215–254.
|
- Andrew R. Golding and Paul S. Rosenbloom.
The evaluation of Anapron: A case study in evaluating a
case-based system.
In Working Notes of the AAAI-94 Workshop on
Case-Based Reasoning, Seattle, 1994.
|
|
| |
Received Gary K. Poock Editorial Award.
|
|
| |
Received Best Paper Award.
|
- Andrew R. Golding and Paul S. Rosenbloom.
Combining analytical and similarity-based CBR.
In Proc. Case-Based Reasoning Workshop,
Pensacola, 1989.
|
Word morphology
|
|
The work on word morphology
involved learning rules of word inflection
in order to store dictionaries more compactly.
This was for my MPhil degree.
|
- Andrew R. Golding and Henry S. Thompson.
A morphology component for language programs.
Linguistics 23 (1985), 263–284.
|
Other
|
- Andrew R. Golding and Neal Lesh.
Indoor navigation using a diverse set
of cheap, wearable sensors.
In Proc. Third International Symposium
on Wearable Computers, San Francisco, 1999.
|
- Andrew R. Golding.
A Review of Case-Based Reasoning.
AI Magazine, Vol. 16, No. 2, Summer 1995.
|
- Rich, Waters, Strohecker, Schabes, Freeman, Torrance,
Golding, and Roth.
Demonstration of an interactive multimedia environment.
IEEE Computer, Vol. 27, No. 12, Dec 1994.
|
- Rich, Waters, Schabes, Freeman, Torrance, Golding, and Roth.
An animated on-line community with artificial agents.
IEEE MultiMedia, Vol. 1, No. 4, Winter 1994.
|
- Andrew Golding, Paul S. Rosenbloom, and John E. Laird.
Learning general search control from outside guidance.
In Proc. IJCAI-87, Milan, 1987.
|
- Steier, Laird, Newell, Rosenbloom, Flynn, Golding, Polk,
Shivers, Unruh, and Yost.
Varieties of learning in Soar: 1987.
In Proc. Fourth International Machine Learning Workshop,
Irvine, 1987.
|
- Rosenbloom, Laird, Newell, Golding, and Unruh.
Current research on learning in Soar.
In Machine Learning: A guide to current research,
eds. Mitchell, Carbonell, and Michalski.
Boston: Kluwer, 1986.
|
Patents
|
Internet
- Andrew R. Golding.
Returning databases as search results.
Pending.
|
- Andrew R. Golding, Michael J. Witbrock, and Alden DoRosario.
Predicting the popularity of a text-based object.
Pending.
|
- Andrew R. Golding and Douglas H. Beeferman.
Estimating the usefulness of an item
in a collection of information.
Pending.
|
- Douglas H. Beeferman, Andrew R. Golding, and
Michael J. Witbrock.
Method and system for collecting related queries.
Pending.
|
Context-sensitive spelling correction
- Andrew R. Golding and Dan Roth.
System for text correction
adaptive to the text being corrected.
US Patent #5,956,739.
|
- Andrew Golding.
System for spelling correction in which the context
of a target word in a sentence is utilized to determine
which of several possible words was intended.
US Patent #5,659,771.
|
English Writer's Assistant
- Yves Schabes, Andrew Golding, and Emmanuel Roche.
Context-based system for accessing dictionary entries.
US Patent #5,845,306.
|
- Yves Schabes, Emmanuel Roche, and Andrew Golding.
System for correcting grammar based on part-of-speech
probabilities.
US Patent #5,537,317.
|
- Emmanuel Roche, Andrew Golding, and Yves Schabes.
System for correcting auxiliary verb sequences.
US Patent #5,535,121.
|
- Emmanuel Roche, Yves Schabes, and Andrew Golding.
Word inflection correction system.
US Patent #5,521,816.
|
- Andrew Golding, Yves Schabes, and Emmanuel Roche.
System for underlying spelling recovery.
US Patent #5,485,372.
|
- Andrew Golding, Yves Schabes, and Emmanuel Roche.
System for correcting improper determiners.
US Patent #5,477,448.
|
Other
- Andrew R. Golding and Neal Lesh.
Indoor Navigation with Wearable Passive Sensors.
US Patent #6,323,807.
|
- Andrew R. Golding.
Adaptive electronic phrase book.
US Patent #6,192,332.
|
- Andrew R. Golding.
Automobile navigation system with dynamic traffic data.
US Patent #5,933,100.
| |
Featured on New England Cable News, evening edition, Sep 30, 1999.
|
Professional activities
Program committees
|
AAAI-94, IMLW-95, AAAI-96, AAAI-98, ICCBR-99, ACL-99,
ICCBR-01, ICCBR-03.
Organizing committees
- ACL-97 workshop: From Research to Commercial Applications:
Making NLP Technology Work in Practice
- AAAI-98 workshop: CBR Integrations
Thesis committee
|
Helen Meng (PhD 1995, EECS Department, MIT)
Organizer
|
4th New England Case-Based Reasoning Roundtable, Cambridge, 1992
Reviews
|
|
Artificial Intelligence,
Machine Learning,
Computational Linguistics,
Language and Speech,
ACM Transactions on Information Systems,
IEEE Transactions on Pattern Analysis and Machine Intelligence,
Applied Natural Language Processing conference,
Siggraph,
NSF,
Oxford University Press.
|
Skills
Programming
|
C, C-shell, Gawk, Perl, JavaScript,
HTML, Lisp, Java, Tcl/Tk
Languages
|
French, some Spanish
Interests
|
|
Bicycling, weight training, guitar, cooking
|
| | |
| |
|
|
