Dear Principal Kehoe.

aspiringtokAI and Robotics

Oct 15, 2013 (3 years and 11 months ago)

72 views

Dear Principal Kehoe.

It was nice to talk to you on the phone today.

Allow me to reintroduce myself.

I am a professor of computer science at the University of California
-
Riverside [a]. I am a past pupil of
Harold's Cross National School (72
-
80), as are my eight siblings, my nieces and my grand niece Jodi.

I am currently working on a project to use comput
ers to help in the fight against malaria (you can read
about it here). A RTE producer, Elizabeth O'Neill who works for Marian Finucane (Elizabeth is CCed to this
email), is making a radio documentary about malaria and will do a little section on my work.


As it happens, I always thought it would be fun to visit my old school and talk to the children, and
Elizabeth thinks it would be interesting to record it for some sound bites. I should mention in passing that
as part of my outreach efforts in science, I h
ave regularly given talks to children as young as eight, on
four continents (but not yet Europe!)

I envision talking to the oldest class, for perhaps 30 minutes. An outline might be something like this..

1) Introduction, my history at

Harold's Cross Natio
nal School.

2) What is a scientist? What is a computer scientist?

3) Using computers to understand anthropology [c] (demonstration of using computers to classify
arrowhead ) I will bring a collection of 6,000 year old arrowheads, and leave them at the scho
ol when I
am done.

4) Using computers to combat malaria (demonstration of classification of insects using my sensor).

5) Brief motivational talk, why math, science and English are so important to study.



If you like, I can let you have a copy of all my sl
ides in advance. I will provide my own computer and
projector. The only thing you need to provide is a projector screen, or a large white wall (ideally out of
direct sunlight).



Please let me know if you have any additional concerns or questions.



Best w
ishes, eamonn





[a] Dr. Keogh’s research interests are in data mining, machine learning and information retrieval. He has
published more than 130 papers, several of which papers have won “best paper” awards. In addition he
has won several teaching awards
. He is the recipient of a 5
-
year NSF Career Award for “
Efficient
Discovery of Previously Unknown Patterns and Relationships in Massive Time Series Databases
” and
several other large NSF and DARPA grants. Dr Keogh has given well received tutorials on time
series,
machine learning and data mining all over the world, and his papers have been referenced well over
9,000 times.

[b]


A Computer Scientist Works to Get the Real Bugs Out


David Zentz for The Chronicle

Eamonn

Keogh, a computer scientist at the U. of California at Riverside, is developing a device to help eradicate malaria
-
causing mosquitoes.


By Tom Bartlett

Riverside, Calif.

In a spare office, using mostly stuff he built in his garage, Eamonn

Keogh is creating a laser
-
powered
device that will save the world. From mosquitoes.

That's the plan, anyway. Right now the contraption is a mess of wires, PVC pipe, and assorted gizmos
connected to clear
-
plastic storage containers. A shipment of mosquitoe
s will arrive in a few days, and
Mr. Keogh, a professor of computer science at the University of California at Riverside, is testing and
fine
-
tuning his creation. The pet project, he admits, has become a full
-
time obsession: It's what he
thinks about while

biking to work, walking the dog, taking a shower. It's never not on his mind.

Mr. Keogh's idea is to build small, cheap sensors that can identify and count mosquitoes, specifically
those that carry malaria, a disease that kills a million people a year. If

the idea works, it will help
researchers determine whether efforts to control mosquito populations are effective and perhaps
lead to new ways of wiping out the pests. (Mr. Keogh is toying with one such method already.) The
idea is promising enough that th
e Bill & Melinda Gates Foundation recently gave him a $100,000
grant to get it off the ground.

Mulling over mosquitoes might seem an odd way for a computer scientist to spend his time, but
almost nothing about Mr. Keogh's career has been typical.

He grew u
p in Dublin, the youngest of nine. His father worked for Guinness, the beer company, and
they lived on a street of other Guinness employees and their families. Never much of a student

he
claims not to have studied even once

Mr. Keogh left school at 15 to b
ecome a car painter. He might
still have paint beneath his fingernails if he hadn't won a visa to come to the United States. When he
arrived, in 1987, he got a job working on expensive antique cars for wealthy collectors.



David Zentz for The Chronicle


David Zentz for The Chronicle

It's never not on his mind: The U. of California at Riverside's Eamonn Keogh thinks about his invention all the time, even in

the shower.

That ended when his boss, who had fallen into debt, shot himself. Not sure what else to do, Mr.
Keogh enrolled at MiraCosta College, near San Diego, neglecting to mention to administrators that
he didn't exactly have a high
-
school diploma. Those were lean
times, and he recalls once having to
decide between buying textbooks and going to the dentist. He chose books and pulled the tooth
himself.

He later transferred to California State University at San Marcos, where he completed his bachelor's
degree in compu
ter science, then enrolled in graduate school at the University of California at Irvine.
Along the way he found he had a passion for computer science, in part because he liked the sense of
control that was missing from his car
-
painting gigs. You can progra
m a computer to do precisely
what you tell it.

Not everyone, though, thought he had the intellectual chops. One professor even took him aside and
warned that he lacked what it took to get a Ph.D.

It's true, Mr. Keogh will concede, that his abbreviated earl
y schooling left some gaps in his formal
education. These days his students can usually best him in math. But Mr. Keogh did finish his
doctorate, and since arriving at Riverside, he's published at a fast clip, achieved tenure, and been
named a university s
cholar. As it happens, he has now published more than the senior professor who
advised him to abandon his ambitions.

Finding Patterns

Mr. Keogh specializes in data mining

which is, basically, looking for patterns in enormous
collections of data. He's best
known among computer scientists for a method of analyzing time
-
series
data called Symbolic Aggregate Approximation, or SAX. So, for instance, you might have a line graph
that shows the highs and lows of a politician's popularity over a year. With Mr. Keogh
's method,
those fluctuations can be converted into simple symbols that might reveal previously unseen
patterns.

The SAX method is widely cited, and Mr. Keogh has written numerous papers on it. He has a puckish
sense of humor, related no doubt to his fondn
ess for writers like P.G. Wodehouse and Flann O'Brien,
so the papers have titles like "Hot SAX," "Experiencing SAX," and "Group SAX." Some of his other
ideas for mildly risqué titles have been shot down by co
-
authors.

While many of his fellow data miners a
re focused on Facebook and Twitter, Mr. Keogh has taken on
less
-
trendy subjects. He is, for example, attempting to catalog and categorize the million or so
arrowheads owned by his university. He hopes to add further evidence to the theory that stone
-
tool
t
echnology evolved in ways that are similar to biological organisms, with techniques being passed
between tribes like bits of genetic material.

Also, as Google busies itself making the libraries of the world available on the Internet, Mr. Keogh is
trying to

come up with ways to make the illustrations in those books searchable, too. The charming
title of his recent paper on that topic: "Mother Fugger: Mining Historical Manuscripts With Local
Color Patches." (The Fuggers were a prominent, powerful European fam
ily in the 14th, 15th, and 16th
centuries.)

Lately, though, he's been preoccupied with mosquitoes. He became interested after learning that
entomologists track mosquito populations with glorified flypaper, sticky traps that require someone
to count, one by

one, the dead insects. That approach seemed ridiculously inefficient and fraught
with the possibility of human error. He started working on a way to automate the process and came
up with the laser idea.

It works like this: The laser is directed at a refle
ctor. The reflected light is then captured by a device
called a photodiode, which can convert light into voltage. When an insect passes through the laser,
the device measures the change in voltage and records it

with a precision fine enough to discern a
wi
ngbeat.

A housefly will beat its wings 190 times per second; a honeybee, 250. A typical male mosquito has a
per
-
second wingbeat total of 600, while the female mosquito flaps less, around 400. The difference
matters because it's only female mosquitoes of ce
rtain varieties that drink blood and, therefore,
transmit malaria.

Mr. Keogh's isn't the only laser
-
related mosquito idea. Last year, at the annual TED conference, an
elite gathering of some of the top minds in the world, the inventor Nathan Myhrvold unvei
led a laser
that he wants to use to actually kill mosquitoes. It zeroes in on the vicious pests and blasts them with
a beam of high
-
intensity light. Whether such a device would be cheap enough to deploy in Africa is
unclear.

For Mr. Keogh, lasers are simpl
y a way of identifying and tracking mosquitoes. However, that may
make them much easier to get rid of. He has already started work on what he calls "the world's
simplest robot," a device that would work in tandem with the sensors he's building. Insects wou
ld be
sucked into a tube and identified by a laser. The harmless ones would then be released, while
malaria
-
causing mosquitoes could be diverted to a canister.

Keeping It Cheap

For any of this to be viable, though, it has to be cheap. If the sensors cost $
1,000 each, or even $100,
then Mr. Keogh's plan of placing them all over malaria
-
plagued regions of the world would be a bust.
For the prototype, which he manufactured himself with the help of a postdoctoral student, he used a
laser pointer from a 99
-
cent
store. He believes he can keep the cost of the sensor under $10.

Still, a multitude of issues remains. There are more than 3,000 kinds of mosquitoes in the world, but
only a few dozen carry malaria. His device would need to know the difference.

That's why
he's talking with entomologists about mosquito behavior in hopes of finding new ways to
track them. For example, some types of mosquitoes are active only in the evening, while others buzz
throughout the day. If he can record when mosquitoes pass through th
e laser field, that might be
useful in making a positive ID.

After mosquitoes, Mr. Keogh plans to turn his attention to cellphones. He has an idea for helping
people in impoverished countries by taking advantage of their phones' built
-
in cameras. The
notion
is still at the proposal stage, so he won't release the details.

It's worth noting, though, that Mr. Keogh does not own, and has never used, a cellphone. In fact, he
doesn't have a phone of any kind at home. It can be inconvenient at times, but Mr.
Keogh says going
phoneless has its upside: No one can bother him while he's thinking.


NGS/Waitt grantee Eamonn J. Keogh and his team are attempting to digitally archive images of rock art, or
petroglyphs, found in the Southwest United States and making t
hese available to the public.


An abstract (top) and animal petroglyph (bottom) which has been traced and issued as queries to the
database that NGS/Waitt grantee Eamon Keogh and his team are building. The list of the five nearest
matches to each are shown

left to right.

Image courtesy of Eamonn Keogh

Petroglyphs are images created by removing part of a rock surface by incising, pecking, carving, and abrading.
Petroglyphs

are one of the earliest expressions of abstract thinking and are considered a hallmark of humanity.
Beyond their value as an aesthetic expression, petroglyphs provide a rich body of information on several different
dimensions. Motifs can, at least in theo
ry, be identified and traced through time and space, which in turn may shed
light on the dynamic histories of human populations, the patterns of their migrations and interactions, and even
continuities with present indigenous societies.

Studies of rock art

have implications beyond anthropology and history. For example, a recent study postulates the
existence of a now
-
extinct Australian bat species based on extraordinarily detailed pictographs known to be at least
17,500 years old. Petroglyphs have been used

in studies of climate change, and the changing inventories of species
in the Dampier Archipelago from the Pleistocene to the early Holocene periods have been reconstructed partly using
petroglyph evidence. However, in spite of these successes, progress in

petroglyph research has been frustratingly
slow.


Clustering of Southwestern USA rock art


Image courtesy of Eamonn Keogh

This could be because the extraordinarily diverse and complex structure of rock art images defies most existing
image matching algor
ithms. Most approaches are simply not suitable for capturing the similarity of petroglyphs, and
those that are, even in limited cases. In this work the team will introduce a novel distance measure for rock art,
showing that it can correctly capture the sub
jective (and where available, objective) similarity between petroglyphs.
They hope to show how they can use this distance measure as a basis for several higher
-
level “data
-
mining”
algorithms (e.g., finding repeated motifs, clustering, or simply enabling qu
ery
-
by
-
content).

In this work Keogh and his team will identify the reasons for this, and introduce a novel distance measure and
algorithms, both of which allow efficient and effective data mining of large collections of rock art.


--


Dr. Eamonn Keogh

Department of Computer Science and Engineering

EB II

University of California Riverside

Riverside, CA 92521
-
0144

http://www.cs.ucr.edu/~eamonn/