First Interdisciplinary Symposium 2015-2016: Natural Sciences

Speakers:

• Michiel van den Broeke, Marine and Atmospheric Research, GRACE.
• Coert van Gemeren, Interaction Technology.
• Laura Filion, Soft Condensed Matter and Biophysics.

Report 1: Hylke Kortenbosch

The central topic of the symposium “Natural sciences” is quite obvious for an interdisciplinary beta-symposium and was clearly reflected in the three talks of the evening. Despite being focussed mainly on physics and computer science the talks were accessible and interesting for a broader public. Which is why the title/topic “ Natural sciences” can be considered a justifiable choice.

The first talk was given by Michiel van den Broeke about the study on the melting Greenland’s icecap. This study focused on three aspects of the melting ice. First, the correlation between temperature and atmospheric composition (CO₂) during the last thousands of years. Data showing this was acquired though ice cores drilled at the Antarctic. Second, tracking and modelling the melting of Greenland’s icecaps both in the present and recent past by using satellites and local measurements on the thickness and weight of the ice. It appears that right now the Greenland icecaps lose 35% more ice during summer than they acquire during winter. Finally, the measurements were used to predict the future of the ice. However, one large unknown remains, future human CO₂ emissions.

The next talk, given by Coert van Gemeren, was about object recognition, otherwise called an engineering approach to cognitive science. This subfield of computer vision is effectively all about teaching a computer to recognize objects in pictures like a human would. This is done by combining convolution, to detect the edges of objects, with neural networking , deep learning. This teaches the computer to recognise patterns within matrixes (filters) on an image and match this pattern to images of an object with that same pattern. This is already quite successful when applied to non-moving images. But when movements enter the equation this is still too complicated, which was ironically illustrated by a video concluding the talk.

After this second talk we all enjoyed a nice dinner at the Locke building next to where the symposium started.

The final talk was given by Laura Filion and was about what physicists do and do not yet know about hard sphere nucleation. Hard sphere nucleation is a process that takes places  with nanometre to micrometre sized, solid particles, colloids, that are suspended in a liquid. Through entropy these object tend to cluster and form crystals when the pressure within the system increases. Even though both the simulations and experiments that are used to study this process seem to be consistent among themselves there is a significant discrepancy between the two approaches. Solving what causes these discrepancies will help us better understand and control the behaviour of colloids in a fluid. And with this, the symposium was concluded.

Report 2: Lotte Pronk

With a broad theme like ‘Natural sciences’ and a programme with speakers from very different research fields, the evening promised to be very interesting for everybody in the audience.

The melting of the ice on Greenland, talk by Michiel van den Broeke
Nowadays CO2-levels are 35% higher than they have been for the last million years, and global temperature has risen by 1 C⁰ over the last 100 years. This increase might lead to the melting of the ice sheets, which causes the sea-level to rise. As we all know, this will cause problems for humans inhabiting the coastal areas. With the “GRACE” project, mass trends of the earth’s surface are being examined with the help of satellites. The satellites have measured a large mass loss in Greenland, which means that the ice sheets are shrinking.  But is this mass loss just part of a natural oscillating trend, or is it anomalous when compared to the past? GRACE has only been operating for about 10 years, so it could be that if we measure for a longer time period, we will see an oscillation. To answer these questions, models and observations need to be combined. There are 40 weather stations that on the ice sheet of Greenland that measure its mass loss. Because this is not enough for a complete picture, a numerical climate model is used to complement the data from the weather stations. This creates a map of the mass loss of the Greenland ice sheet.
The data show that during the last 10 years, the mass gain (by snowfall) does not balance the mass loss (by iceberg formation and meltwater drainage) anymore, which leads to a net loss of mass.
According to Michiel van den Broeke, this means that the ice sheet is out of balance and that the large mass loss that is measured by GRACE is indeed an anomaly, and not part of a naturally occurring oscillation. So what about the future? That depends on what we will do: In order to prevent dramatic rising of the sea level, we need to use the results from this and other research to adjust policies and decrease our CO2 emissions.

Recent advances in object recognition, talk by Coert van Gemeren
For computers to recognise objects, statistics and machine learning need to be combined. Objects can be identified, classified and detected. The detection of objects in a picture with all kinds of background is very difficult, because you need to give a meaning to RGB-values.  Convolution is based on multiplying (in a special manner) two matrices: an image patch (RGB-values) and a kernel. The outcome creates a new image, which is used for detection.  Since objects can come in many variations, the static kernels are not really suited for detection.  A solution is to use a Convolutional neural network (CNN), which is inspired by biological processes, such as the visual mechanisms in our brain. CNN is a form of “deep learning”, which you might have heard of before. After an explanation on how CNN works, he gave a demonstration in which a picture of a jaguar was recognised by the neural network.  Concluding, CNN shows great promise for object recognition, but there are still some downsides:  CNN need to be trained, for which a lot of data is needed. It thus takes time and computer power to do this. Furthermore, 3D-recognition and object recognition of movement is not possible with CNN.
A nice demonstration of what will happen if you apply CNN on a video can be found here:  https://www.youtube.com/watch?v=DgPaCWJL7XI

Dinner
After these two interesting talks, we were led to another building for an Indonesian dinner. Everybody took a seat in one of the rooms that were available, and enjoyed the lovely food while discussing the topics of the talks or sharing the latest gossip.
Apparently the Auditorium was double-booked, but the committee handled this situation very well and the last talk of the evening could still take place in an available lecture hall.

Hard sphere nucleation, talk by Laura Filion
This talk was about physics. About hard sphere nucleation to be precise. Fortunately, the talk was very understandable, even for non-physicists. Laura Filion researches the crystallyzation of hard spheres: model particles that are similar to colloids. Colloids are nanometer to micrometer size that, in a solvent, undergo Brownian motion and that can self-assemble into different phases.
Experiments show that hard spheres do crystallize, but it is difficult to simulate this. Some simulation methods were explained, and all these methods give similar results. But the results from the simulations are not in line with experimental results from other research groups. So who is right? Obviously more research is needed to answer this question. In the future, confocal microscopy might help to solve this problem.
All in all, this symposium was interesting for everyone in the audience because of the broad range of the subjects. It was kind of disappointing that one of the speakers did not show up, but the other talks were interesting enough to make it an enjoyable evening!

Photo report of the natural sciences symposium: Anne Snijders

(click to open)