In vivo, in vitro... in silico!

Virtual reality has invaded our daily lives, not to mention that of the researcher. Ranging from storing nuclear waste to computer vision, digital simulation has become one of the pillars of science in the XXIst century. Even farmers are not going to be outdone and will soon have real "digital" beetroot!

For scientists, resorting to computer simulations to describe phenomenon has become, in certain areas, something that they cannot do without. Take astrophysics for instance. Researchers in this area are unable to carry out experiments due to the gigantic size of the objects they study. They have no other choice but to turn to digital replicas of the stars and the galaxies if they want to decode the mysteries of outer space. As for medical researchers who wish to cure heart pathology, how can they study the effects of defibrillation on the human body without the binary version of this vital organ? After having been in vivo and in vitro, medicine has become “in silico”, according to the expression used in laboratories to describe research that relies on digital simulations (the word silico refers to the silicon, material made up of microprocessors).

"Before, science was based on comparing theory and experimentation. Now, it is based on triptych theory, experimentation and digital simulations”, summarizes Michel Kern, who is in charge of research at Inria (http://www.inria.fr/index.en.html the French national institute for research in computer science and control) within the Estime team (Parameter estimation and modeling in heterogeneous media). Based in Rocquencourt, he works in one of institute’s eight research centers whose purpose is to carry out fundamental and applied research in the areas of computer and communication science and technology.

What is digital simulation and how can it be obtained? Michel Kern is going to teach us using a simulation from the underground research laboratory of Bure in the Meuse (part 1). Preliminary step to simulation, modeling. This involves presenting a state or a situation in the form of equations. The model is then used to simulate, as closely as possible, the changes in the situation in response to different parameters (weather, environment, etc.) But how can these computer simulations help those in the civil sector? Paul-Henry Cournède, researcher in “virtual agronomy” answers this question (part 2). And finally, Olivier Faugeras, specialist in computer vision at Inria, will show us how these simulations may be able to help scientists to better understand how the human brain works (part 3).

01.Escaping from the constraints of “time”

What archetypal discipline needs digital simulation? Geology; a world where everything moves slowly. Everything happens over a timeframe of hundreds of thousands of years, if not more. Researchers on the Estime team are trying to reduce these lengths of time. At Inria, they are carrying out research on the movement of fluids in the subsoil. They are studying, in particular, the problems connected to the burying of nuclear waste on a possible storage site near the underground research laboratory at Bure.

Indeed, the behavior of the material used for storage leads them to believe that the equivalent of one thimble of radioactive fluid will be released from the site every year. The storage site was dug 500m deep, in an impermeable layer of clay and it will not take long for the fluids to cross the surrounding rock and contaminate the nearby water tables. In the meantime, will they have lost their radioactivity? In an attempt to answer this question, Andra (the national radioactive waste management agency) in charge of this site, is carrying out digital simulations in order to evaluate the security of the storage site over a million years. As specialists in the simulation of the flow of every type of fluids (water, hydrocarbon, liquid Co2) in the substratum, the Estime researchers are lending Andra a hand in this domain.

What does the story-board for the work of simulation look like? It all starts at the library, with a phase involving bibliographical research. “In the eighties, scientists came to see mathematicians to say:I have this integral, can you calculate it for me?” explains Michel Kern, member of Estime. “These days are over. It has become more and more important to have good knowledge of the area in which you are carrying out simulations,” in this case a background in hydrodynamics and geochemistry. This often involves scientists from different horizons having to sit down together; even if this means being faced with misunderstandings between the different participants. “For example we don’t always speak the same language as biochemists, nor necessarily have the same objectives. A major part of this work is to manage to understand each other.” From this exchange and the work of compiling a bibliography, the researchers have illustrated the phenomenon of the flow of fluids in the form of a “model”, in other words, a set of mathematical equations. This is the first step, called modeling.

The researcher must then go on to creating a “digital” version of the natural environment that is the object of the study. Digitalizing the environment involves redrawing it using geometrical objects to obtain a virtual image. The exact shape of these objects is not neutral. “The first computer program that we wrote did not allow us to use deformed hexahedrons”, recalls Michel Kern. “And yet we needed this deformation to simulate the rock bending in places.” In fact, we know that clay, when containing water, will deform over time. “Therefore, we had to revise the program in order to be able to take into account deformed hexahedrons.

Final step: design and implement the algorithm that will simulate the scientific model, taking into account the natural environment. Algorithms are a sequence of instructions to obtain a given result. For example, when you are looking up a word in a dictionary, you go about it in a logical and systematic way, by looking at the letters, turning the pages, etc. When you put these entire procedures one after the other, it forms an algorithm. Algorithms are not always a sequence of mathematical operations but in our case, they are. The algorithm of the Estime team allows us to evolve the scientific model over time and to obtain a simulation of the flow of radioactive fluid over the long term.

The difficulty in designing an algorithm lies in the astronomical number of equations at stake. Year after year, researchers improve these algorithms and efficiency has improved by ten million since the 60s and 70s. By improving these, modelers are like astronauts equipping telescopes with more powerful telescopes: they can see further ahead in time.

02.Creating “digitally modified” organisms

Today, sectors in engineering that do not use simulation are few. Thus, the RATP (the Parisian transport network) calculates the evacuation time of stations in the case of a fire using software that models panic movements. In addition, in May 2008 Microsoft launched a service that forecasts traffic jams in some American towns using a traffic jam simulator. As for the gem of European aeronautics, the A380, it would simply not exist without digital simulation.

The agricultural sector could also be added to this list in the future, thanks to the Digiplante team at Inria. Its researchers use digital models of plants to optimize agricultural yield. What does a digital model of a plant look like? For the user it looks like a plant drawn on a computer screen. For its creator, it corresponds to about ten parameters that determine approximately how the plant is going to grow under certain environmental conditions. In this way, researchers will be able to study how the plant will grow in one area or another. They simulate the growth by modifying, via the digital version, conditions in the natural environment (rainfall, sunshine, quantity of nitrogen fertilizer, type of soil, etc.). By applying this recipe to sunflowers in the region of Montpellier, Digiplante has been able to show how the production yield of sunflowers could be increased by 30% by better dispersion of water throughout the year!

Apart from the sunflower, the team also has a digital model of the beetroot that should soon be used to optimize French vegetable crops. “We have worked out an “ideotype”, that is to say an ideal organism for a given geographical situation. Now, it is up to the geneticists to determine what genes would allow these ideotypes to be obtained,” explains Paul-Henry Cournède, researcher at the Ecole Centrale of Paris and member of the Digiplante team, spread over several sites. Concerning this genetic research, the work has already started at the ITB (beetroot technical institute), a private body that has the duel role of carrying out experiments and distributing the information about technical progress to farmers, which could lead to improving production and the quality of their crops. In the future, the beetroot we eat could well have been “digitally modified”.

Although the expected results are simulation tools, tools to optimize and control agriculture, forestry and ecology, for now, Digiplant’s digital models are essentially used in plant visualization software. These virtual plants are used, for example, to improve the realism of French army flight simulators or to decorate Ubisoft’s video games. Architect's offices also embellish their models of buildings by using credible vegetation.

03.In pursuit of animal vision

What would a vision simulator look like? To a devise that could pick up an image and “mime” neuron reactions. Olivier Faugeras, science academician and Inria research director, is trying to get this devise up and running. He has just won a call for tender at the European Council of Research for innovative projects in applied mathematics. This endowment of €2m will enable him to create a simulator of animal sight.

Where does the challenge lie in all this? In any case it is not a question of the power of calculation. The collective power of a hundred billion neurons of the brain is around ten elementary peta instructions per second (one peta= 1015). Or approximately the calculation power of the most powerful computer on the planet, Roadrunner, located in the laboratory premises in Los Almos in the USA and that turns at five peta instructions per second. Indeed, the neuronal power would soon be accessible for Roadrunner and other computers, given that the power of these machines is doubled every eighteen months. The problem therefore lies in the software design rather than its performance. What approach should be taken to mime the brain?

"Researchers that try to model all the workings of all the neurons at the same time are heading for failure”, says Olivier Faugeras. In concrete terms, the world computer vision specialist is not going to design one, but several digital models. Each model will represent a different level of complexity, ranging from the functioning of an individual neuron to that of a network. This network could, for example, cover the hundreds or cortical areas involved in sight and specialized in tasks such as recognizing a face or a shape. The models will be interlinked to each other like a set of Russian dolls.

One of the difficulties is to find a mathematical description of the interdependence between the cortical areas. These zones are in constant interaction. “In order to map the complexity of behavior in these zones we will use the functional analysis and the bifurcation theory,” the researcher plans. Functional analysis is one branch of mathematics that enables us to analyze general equations based on functions rather than, more classically, on numbers. In this case the functions represent “mathematical operations” carried out on the cortical areas. For Olivier Faugeras, the objective is to build a devise capable of making the different cortical area models communicate with each other. The bifurcation theory describes how, with slight changes in its characteristics, a system that can totally change its behavior. In other words, for a vision simulator, how the interdependence between the cortical areas can evolve towards a chaotic regime.

Once the model is obtained, Olivier Faugeras and his team will put it to the test. They will use a classical battery of brain investigation techniques: functional MRI, magneto encephalography and electroencephalography. By showing to guinea pigs (humans or monkeys), and to the simulator itself, films depicting scenery, faces, etc., they can check that the consecutive changes in the cortical regions agree with the model’s forecasts.

Apart from the theoretical challenge that it represents, this project is also a technological test, as Olivier Faugeras points out: “We want to compare the performance of a system that has been inspired by animal vision with current algorithms of visual processing. The best image analyzer today is still our eye and our brain”. Today computer vision algorithms are used every time image processing is necessary: when analyzing medical images, when indexing photos or videos for archiving purposes, or in robotics to help rover explorers on Mars find their way around. Will Olivier Faugeras’ software compete with classical algorithms used in application programs? Make a date in five to eight years to know if this project has lived up to its (great) expectations.

Add comment

Add comment
Validation key

To validate the form, enter the first 4 characters and the last 1 character in the series.

  • A
  • 5
  • 4
  • M
  • H
  • 9
  • U
  • C

*Information Requiered

Stay connected

Follow us : Flux RSS Facebook Page Twitter Page

Newsletter :

Junior

Claude Lévi-Strauss: first encounter with the Indians

Claude Lévi-Strauss: first encounter with the Indians

This dean of the Académie Française, where he was elected in 1973, and great ethnologist celebrated his 100th birthday on the 28th November of 2008. In a world that he never really cared for much...

See more

Multimedia

Cette vidéo nécessite le plug-in gratuit Flash 8.
Il semble que vous ne l'avez pas.
Cliquer ici pour le télécharger

Mathematics and folding

  • Video

Valérie Larose teaches mathematics in the Louise Weiss Junior High School in Nozay (Essonne, France). She shows us how she uses folding to demonstrate different mathematical concepts to pupils aged 11 (6th grade class).

Sharing Experience

Junior Conferences at the CEA

Junior Conferences at the CEA

Every trimester, the French National Institute for Nuclear Science and Technology (INSTN), part of the French Atomic Energy Commission (CEA) based in Saclay, France, organises a free conference aimed at young people. Middle and high school students come, sometimes accompanied by their parents, to listen to the researchers describing their work. The topics, often specialized, are always accessible and outlined with concrete examples.

See more