This piece appeared in the March 2007 issue of the International Society for Neuroethology newsletter.
You may reproduce this article only with
proper attribution as follows:
Kaushik Ghose, “Flights into the unknown”, International Society for Neuroethology newsletter, March 2007 issue.
Flights into the unknown
Kaushik Ghose
When
I came to Cynthia (Cindy) Moss' batlab at the University of Maryland,
"the bat" in my mind was an abstract construct. For the previous five
years I had trained in a field where people had spent a century
understanding and distilling empirical knowledge into neat,
analytically tractable portions. I was trained to think of giant,
complex pieces of machinery in terms of simple 'equivalent circuits'. A
hydroelectric turbine - tons of wires and metal scaffolding rotating
thousands of times a minute, converting the kinetic energy of last
year's rainfall into megawatts of electricity - was represented on a
regular sized sheet of paper by a few abstract differential equations
based on a little circuit diagram. It was a powerfully seductive way of
looking at things. The reality was violent and chaotic. The abstraction
was serene and understandable. The abstraction gave the comforting
vision that we understood and controlled some of the powerful forces of
nature and man. As sophomore electrical engineers, however, we had been
shown, for example, that transformers would get hot, would hum and
would sometimes explode. These were behaviors that our little circuit
diagrams could not explain[1]. It had been my first tantalizing glimpse
into the unknown. Into lands that lay beyond the vague and short
boundaries of what we knew and had plowed into equations and diagrams.
When I started my PhD course I wanted to model "the brain".
For my masters I had moved on from circuit diagrams and transformer
design to signal processing, computer vision and pattern recognition.
These engineering problems were always being "solved". They would be
solved for this dataset and that dataset, but they would never be
*solved* in the broad sense. You could write a program to recognize 90%
of a set of handwritten digits, but not 95%. A human, on that set,
would do about 100%. How could a human do 100%? What kind of program
could model the human brain, when, for instance, it was analyzing
speech? That seemed to be a good question to study. I did not know it
at that time, but I had stumbled across one of the biggest mysteries of
our time - a giant continent, largely untravelled and unmapped, full of
mystery and surprise.
The first few months in the batlab I puttered around with
this abstract bat that had an abstract brain with abstract neurons and
lived in a computer (well, mostly in a dog-eared notebook). Cindy
quickly introduced me to Timothy (Timmer) Horiuchi who had just joined
the University of Maryland. Timmer was planning to make robot bats with
brains made of resistors, capacitors and transistors. He called it the
'microchipoptera' project. When you said this was how the brain worked,
and you put it on a robot and you let the robot loose in the hallway,
you were putting your money where your mouth was. When everything sat
inside a computer you could potentially "solve" and "explain" anything.
You were modeling both the brain and the environment it would interact
with. The temptation to model the brain and the environment so that
they fit and worked would be great. And it would possibly lead you to a
tautology: A circular argument hidden under layers of sophisticated
mathematical reasoning and computer code. When you put your model brain
on a robot and let it interact with the real world, there would be less
opportunity to succumb to this philosophical disease. Uptil then, I had
some vague idea of what bats did. I had never seen a real bat before, I
had never done animal experiments, and I was a little afraid of the
fact that bats bit and could carry rabies. And I didn't know if I could
learn to do surgeries. On the other hand, I could code computer
programs and design electrical circuits. I got very excited by the idea
of this robot bat. It seemed something I could do. A science fiction
fan could hardly turn down the opportunity to build a robot bat! It
turned out to be an excuse. I wanted to launch an expedition into
uncharted territory, but I was afraid of dark corners.
Some time into simulating electrical circuits that would be
used to model attention in Timmer's Computational Sensorimotor Systems
Lab I got my rabies pre-exposure shots, which meant that I could go
into the animal area. More importantly I could sit in on behavioral
experiments people were running in the lab. I went in one day when Jeff
Triblehorn was running one of his experiments. Jeff was researching
down the hall in David Yager's insect hearing laboratory. He was
running a collaborative experiment with our lab studying how praying
mantises responded to bat attacks. I went into this darkened room where
Jeff was releasing mantises and video taping bats fly after them. I
watched through the monitor. I watched the slow motion recording as the
bat flapped its wings, as the mantis whirred, as the bat dipped,
weaved, looped and spiraled after it. I watched as it reached out its
wing in mid flight, as it tipped the flying mantis into its tail
membrane, as it grabbed it in its mouth, as it pulled up inches from
the floor. "I want to study THAT", I said. "I want to know THAT, I want
to understand THAT. I want to build a robot model of THAT. I want to
know how neurons do THAT". The bat wasn't an abstract concept anymore.
It wasn't a "model system", a "problem", a "research topic". It was an
animal, a living being that did something, and did it well. And I
wanted to understand THAT. My desire to explore had overcome my fear of
the unknown.
It turns out the University of Maryland is a great place in
pursuit of THAT. In Timmer's lab while I was trying to build a model of
how a bat would use its echolocation system to deal with complex
environments, I ran into many questions whose answers were necessary to
build a useful model. Cindy's lab with its behavioral and
electrophysiological setups was the ideal place to try and answer some
of those questions. I made extensive use of the high speed cameras and
large flight room in the Batlab to study where bats directed their
sonar beam while chasing insects. During this time, another researcher
in the lab, Murat Aytekin, got intensely interested in the question of
how a bat can use echo information to localize objects. In
collaboration with Jonathan Simon over at the department of Electrical
Engineering, Murat and Cindy came up with a general theory of how
animals could use sound alone to develop a sense of space. A pertinent
question, since in bats vision is a low acuity sense, whereas
echolocation seems to be an extremely fine grained sense, spatially.
During the course of my experiments Cindy introduced me to
P.S. Krishnaprasad. PSK studies control systems of both the inanimate
and animate kinds. His particular speciality is in taking complex real
world systems and applying strange and wonderful results from the world
of mathematics to pull out (of the hat as it were) an analytical
understanding of, say fish schooling, bats swarming or people on
uni-cycles. Together, PSK, Timmer, Cindy and I showed mathematically
how the trajectory bats follow when chasing an erratically moving prey
is time-optimal, a useful thing, if you only have a fraction of a
second to catch your lunch.
During this period, as I was eagerly foraging in this
mysterious land of bat behavior I was impressed by how important the
whole community is to one's research. At 11:00am, every Friday, during
the school year, everyone in Maryland interested in the neural basis of
behavior gathers to hear talks by people at Maryland. This diversity of
ears makes for a challenging presentation, since a talk on bat
echolocation should be accessible to people working on molecular
mechanisms of learning in rats (Betsy Quinlan's lab) as well as people
mapping spatio-temporal receptive fields in ferret auditory cortex
(Shihab Shamma's tribe). However, from this diverse and lively group
will come diverse and lively comments on your approach and your
results. I have gotten, on several occasions, questions and comments
that forced me to think outside my box. Jens Heberholz, for example,
who is currently studying social interactions in crayfish, looks at my
results on sonar beam patterns in echolocating bats and thinks of
evolutionary questions. Stephen V. David looks at the same data and
thinks of parallels to the visual system in primates, which he studied
before joining here. Kate McCloud, from Catherine Carr's lab asks
challenging questions about anatomical substrates of sonar beam shape.
Todd Troyer looks at some pursuit data from bat-insect chases and
thinks of patterns of play in American Football. Not only is doing the
science fun, but so is presenting it to such a lateral thinking
audience.
Right, now, as I write this I'm trying to build a model of a
subset of neurons in the inferior colliculus of the bat. People suspect
that these neurons have something to do with sound localizaton in the
bat, but they have some puzzling timing properties. I'm wondering if I
put together a model of these neurons whether their projected
population response will tell me something that is not obvious from
their known single unit properties. But I'm stymied by some conflicting
reports in the literature I'm going through. Perhaps a quick
presentation at one of those 11:00am Friday meetings will help me come
up with some ideas to get unstuck. It might even help me avoid a
certain philosophical disease that afflicts people making computer
models...
[1] These particular phenomena are actually resonably well understood and modeled, unlike the human brain.
(c) Kaushik Ghose 2007