00:11
Alright, so Masako, I have good news today, which is rare.
Yeah, it's rare. Okay, what happened?
Let me show you. Yeah, so I finally got a job after my graduation. So I have my postdoc position.
Congratulations.
Thank you. I honestly didn't think it would be decided so quickly. I thought that
I was fully prepared to not have anything lined up when I graduate and I was going to just,
you know, graduate, chill for a little bit while I job hunt, start after a couple of months later
or something. That's sort of what I had in mind. But yeah, I think it's good to have things sort of,
I know where I'm going after this.
Yeah, so what are you going to do and where would you be?
So yeah, so this is actually a dream job kind of for me. So I have started my PhD with the
desire to work in a museum after PhD as a scientist. And so the typical path for this
kind of job is to go doing postdocs at a museum or do a postdoc with the university's lab that
collaborates heavily with museum. So if I were to go to the museum route, the fellowship
applications are typically around January, February deadline, which is why I started to
look for it. Even though I didn't feel super ready to do the job hunting, I felt like, well,
if I miss this deadline, it will be another year. So, you know, let me just see what happens.
And so I applied to a couple of places, other different museums. But this one came around
through sort of like a Listserv. Do you know what a Listserv is?
Oh, yes.
Yeah, yeah, yeah. So Listserv for conservation scientists stuff. Like while I was looking for
fellowship jobs, I randomly, without thinking too much about it, came across this Listserv
for AIC, American, I don't know what I stands for, but conservation.
American Institute of Conservation, something like that. I should probably check. But I found
a Listserv. They were doing lots of job postings and stuff. And I looked at some of the history
and looked pretty relevant, even though they were mostly looking for conservators,
like people who actually fix the painting or the artwork.
03:12
In those listings, I would see occasionally some science positions. So I decided to just sign up.
And, you know, when you sign up to the Listserv, you get like a weekly digest or like daily,
you know, highlights on the things that you want to hear about. And this job popped up. And I was
really surprised because it's actually the project is going to be building like optical instruments
at the museum. And I, you know, my PhD is in ultra fast molecular dynamics. Part of it
involves designing and implementing optical setup for the lasers.
So that can be used, yeah. How can it be used? Can you?
Yeah, so it's not exactly going to be the same kind of setup, but the basics of, you know,
I'm designing optics so that you can fit a certain dimension of things in a given space or,
you know, lots of different little optical requirements like velocity, group velocity
stuff or other things. You know, chromatic aberration, all these like problems that
you will encounter are similar. And so even though I've never actually built a microscope
before, this is the plan that I will be doing. And I think I'm actually pretty equipped to do this.
And I'm excited to do that in a museum as well. Because that was sort of my goal to be able to
start dabbling into my career in a museum scientist world. And so I'll be doing that.
And so I'll be reporting to three PIs. One PI will be the optical design person.
Other one is from the museum. And the other one is also from university, but he is
a convolutional neural network guy. So here's the part that I am very unconfident about.
This is going to be a new challenge for me. So the first part of the project is me
designing, buying the parts that I need and building this microscope, special kind of
microscope in a museum. Second part of the job is to build the analysis scheme.
06:01
And what they're trying to do is to incorporate convolutional neural network to analyze the
image from the microscope. Because what they're trying to do is basically using
a image recognition, like pattern recognition system to streamline the characterization
process. So in a museum, oftentimes what we want to know is what are things made of,
right? What are these artworks made of? And we want to do it in a way that is least
invasive to the artwork so that we don't accidentally damage the artwork.
And we want to do it accurately. So we need high-res information, but also
a non-invasive mechanism. And this particular type of microscope turns out to be a pretty good
candidate for that. But right now, what they're doing is to use their microscopic images and say,
this looks a lot like this other reference spectrum from the pure sample. Because imagine,
painting, artworks, they're all mixed with a bunch of other things. It's not just the
pigment that's present, right? There's binding medium, there's varnishes, there's
other things involved. But basically, it relies on human eyes to match with a reference spectrum
and say, okay, this looks a lot like a known pigment called this. And
that part was a manual process. So I mean, nothing wrong with that. That has worked out very well for
some time. But it still relies heavily on human judgment to do this. What we're trying to do
is to use convolutional neural network so that we can process lots of images with
hopefully high accuracy. So the inputs will be the data that you will acquire from the
experiments? Like the spectrum? Yes. Like the spectrum data?
Well, not necessarily spectrum. It will be the image. So that will be image data. So rather than
having like a vibrational spectroscopy. So that's sort of the vibrational spectroscopy,
Rydberg spectroscopy, that's sort of been a realm that I have been in my PhD. So it's sort of like
an indirect information. So the data you will get will be like an image itself?
Image. Image itself on the CMOS or some kind of imaging detector. And the idea is to process
09:05
as image while also training the neural network to be able to identify different mixtures
from mixture of the samples. Like can you identify the components of different mixtures?
Or how well can you identify two very similar signals?
So I'm really naive about all of these. So am I.
Can you speculate or estimate like the decay of materials used?
So that I think so adding the time axes onto this information because images are just snapshots.
And I think that will be an interesting sort of future goal to have because one of the main
aspects of conservation science is like how do artworks, paintings, whatever decay over time?
Inevitably they do get damaged over time. Mostly usually by light, photo decay, but not just that.
There's humidity involved, there's all sorts of other things involved. There's actually
very interesting research going on by the team in the University of Amsterdam.
Noticing that there is a metal soap like saponification reaction that's happening
in a time span of like hundreds of years. Slowly developing soap-like molecules in the
in-between region in the painting in a binding medium. And that travels towards the surface of
the painting making the painting like bubble up quite literally. And that damages the integrity
of the surface of the painting which is very important, right? So that's sort of like dynamic
kinetic studies are definitely very interesting for me as well. But the one thing is a lot of
the times these are very slow reactions. So I'm not sure how like I'm not super familiar with this
artificial aging process. People do that in museum science like they put it in like a chamber with
like exposed you know they measure the flux of light and you know estimate okay this is roughly
however many years, days of light exposure or something like that. You could do that and there
are people who are doing studies with that. I'm not sure how compatible it is
to do this in an imaging technique. But I think that would be definitely interesting especially
for something that we know very well like oh pigment A decays to A star and then to A star star
12:04
or something like that. And if we know the pathway really well then maybe capturing that is not going
to be super duper difficult. But first thing first is kind of just hey can we identify things
well and if we can rely on neural network and we can all share the database with the rest of the
world that would be very very fruitful and meaningful project. Sounds exciting. Yeah I'm
super excited about it. Okay so that's it for the show today. Thanks for listening and find us
at Eigo de Science on Twitter that is E-I-G-O-D-E-S-C-I-E-N-C-E. See you next time!
