New Algorithm Offers Ability to Influence Systems Such as Living Cells or Social Networks

However, an MIT researcher has come up with a new computational model that can analyze any type of complex network -- biological, social or electronic -- and reveal the critical points that can be used to control the entire system.

Potential applications of this work, which appears as the cover story in the May 12 issue ofNature, include reprogramming adult cells and identifying new drug targets, says study author Jean-Jacques Slotine, an MIT professor of mechanical engineering and brain and cognitive sciences.

Slotine and his co-authors applied their model to dozens of real-life networks, including cell-phone networks, social networks, the networks that control gene expression in cells and the neuronal network of the C. elegans worm. For each, they calculated the percentage of points that need to be controlled in order to gain control of the entire system.

For sparse networks such as gene regulatory networks, they found the number is high, around 80 percent. For dense networks -- such as neuronal networks -- it's more like 10 percent.

The paper, a collaboration with Albert-Laszlo Barabasi and Yang-Yu Liu of Northeastern University, builds on more than half a century of research in the field of control theory.

Control theory -- the study of how to govern the behavior of dynamic systems -- has guided the development of airplanes, robots, cars and electronics. The principles of control theory allow engineers to design feedback loops that monitor input and output of a system and adjust accordingly. One example is the cruise control system in a car.

However, while commonly used in engineering, control theory has been applied only intermittently to complex, self-assembling networks such as living cells or the Internet, Slotine says. Control research on large networks has been concerned mostly with questions of synchronization, he says.

In the past 10 years, researchers have learned a great deal about the organization of such networks, in particular their topology -- the patterns of connections between different points, or nodes, in the network. Slotine and his colleagues applied traditional control theory to these recent advances, devising a new model for controlling complex, self-assembling networks.

"The area of control of networks is a very important one, and although much work has been done in this area, there are a number of open problems of outstanding practical significance," says Adilson Motter, associate professor of physics at Northwestern University. The biggest contribution of the paper by Slotine and his colleagues is to identify the type of nodes that need to be targeted in order to control complex networks, says Motter, who was not involved with this research.

The researchers started by devising a new computer algorithm to determine how many nodes in a particular network need to be controlled in order to gain control of the entire network. (Examples of nodes include members of a social network, or single neurons in the brain.)

"The obvious answer is to put input to all of the nodes of the network, and you can, but that's a silly answer," Slotine says."The question is how to find a much smaller set of nodes that allows you to do that."

There are other algorithms that can answer this question, but most of them take far too long -- years, even. The new algorithm quickly tells you both how many points need to be controlled, and where those points -- known as"driver nodes" -- are located.

Next, the researchers figured out what determines the number of driver nodes, which is unique to each network. They found that the number depends on a property called"degree distribution," which describes the number of connections per node.

A higher average degree (meaning the points are densely connected) means fewer nodes are needed to control the entire network. Sparse networks, which have fewer connections, are more difficult to control, as are networks where the node degrees are highly variable.

In future work, Slotine and his collaborators plan to delve further into biological networks, such as those governing metabolism. Figuring out how bacterial metabolic networks are controlled could help biologists identify new targets for antibiotics by determining which points in the network are the most vulnerable.

Source

Robot Engages Novice Computer Scientists

A product of CMU's famed Robotics Institute, Finch was designed specifically to make introductory computer science classes an engaging experience once again.

A white plastic, two-wheeled robot with bird-like features, Finch can quickly be programmed by a novice to say"Hello, World," or do a little dance, or make its beak glow blue in response to cold temperature or some other stimulus. But the simple look of the tabletop robot is deceptive. Based on four years of educational research sponsored by the National Science Foundation, Finch includes a number of features that could keep students busy for a semester or more thinking up new things to do with it.

"Students are more interested and more motivated when they can work with something interactive and create programs that operate in the real world," said Tom Lauwers, who earned his Ph.D. in robotics at CMU in 2010 and is now an instructor in the Robotics Institute's CREATE Lab."We packed Finch with sensors and mechanisms that engage the eyes, the ears -- as many senses as possible."

Lauwers has launched a startup company, BirdBrain Technologies, to produce Finch and now sells them online atwww.finchrobot.comfor$99 each.

"Our vision is to make Finch affordable enough that every student can have one to take home for assignments," said Lauwers, who developed the robot with Illah Nourbakhsh, associate professor of robotics and director of the CREATE Lab. Less than a foot long, Finch easily fits in a backpack and is rugged enough to survive being hauled around and occasionally dropped.

Finch includes temperature and light sensors, a three-axis accelerometer and a bump sensor. It has color-programmable LED lights, a beeper and speakers. With a pencil inserted in its tail, Finch can be used to draw pictures. It can be programmed to be a moving, noise-making alarm clock. It even has uses beyond a robot; its accelerometer enables it to be used as a 3-D mouse to control a computer display.

Robot kits suitable for students as young as 12 are commercially available, but often cost more than the Finch, Lauwers said. What's more, the idea is to use the robot to make computer programming lessons more interesting, not to use precious instructional time to first build a robot.

Finch is a plug-and-play device, so no drivers or other software must be installed beyond what is used in typical computer science courses. Finch connects with and receives power from the computer over a 15-foot USB cable, eliminating batteries and off-loading its computation to the computer. Support for a wide range of programming languages and environments is coming, including graphical languages appropriate for young students. Finch currently can be programmed with the Java and Python languages widely used by educators.

A number of assignments are available on the Finch Robot website to help teachers drop Finch into their lesson plans, and the website allows instructors to upload their own assignments or ideas in return for company-provided incentives. The robot has been classroom-tested at the Community College of Allegheny County, Pa., and by instructors in high school, university and after-school programs.

"Computer science now touches virtually every scientific discipline and is a critical part of most new technologies, yet U.S. universities saw declining enrollments in computer science through most of the past decade," Nourbakhsh said."If Finch can help motivate students to give computer science a try, we think many more students will realize that this is a field that they would enjoy exploring."

Source

Scientists Afflict Computers With 'Schizophrenia' to Better Understand the Human Brain

The researchers used a virtual computer model, or"neural network," to simulate the excessive release of dopamine in the brain. They found that the network recalled memories in a distinctly schizophrenic-like fashion.

Their results were published in April inBiological Psychiatry.

"The hypothesis is that dopamine encodes the importance-the salience-of experience," says Uli Grasemann, a graduate student in the Department of Computer Science at The University of Texas at Austin."When there's too much dopamine, it leads to exaggerated salience, and the brain ends up learning from things that it shouldn't be learning from."

The results bolster a hypothesis known in schizophrenia circles as the hyperlearning hypothesis, which posits that people suffering from schizophrenia have brains that lose the ability to forget or ignore as much as they normally would. Without forgetting, they lose the ability to extract what's meaningful out of the immensity of stimuli the brain encounters. They start making connections that aren't real, or drowning in a sea of so many connections they lose the ability to stitch together any kind of coherent story.

The neural network used by Grasemann and his adviser, Professor Risto Miikkulainen, is called DISCERN. Designed by Miikkulainen, DISCERN is able to learn natural language. In this study it was used to simulate what happens to language as the result of eight different types of neurological dysfunction. The results of the simulations were compared by Ralph Hoffman, professor of psychiatry at the Yale School of Medicine, to what he saw when studying human schizophrenics.

In order to model the process, Grasemann and Miikkulainen began by teaching a series of simple stories to DISCERN. The stories were assimilated into DISCERN's memory in much the way the human brain stores information-not as distinct units, but as statistical relationships of words, sentences, scripts and stories.

"With neural networks, you basically train them by showing them examples, over and over and over again," says Grasemann."Every time you show it an example, you say, if this is the input, then this should be your output, and if this is the input, then that should be your output. You do it again and again thousands of times, and every time it adjusts a little bit more towards doing what you want. In the end, if you do it enough, the network has learned."

In order to model hyperlearning, Grasemann and Miikkulainen ran the system through its paces again, but with one key parameter altered. They simulated an excessive release of dopamine by increasing the system's learning rate-essentially telling it to stop forgetting so much.

"It's an important mechanism to be able to ignore things," says Grasemann."What we found is that if you crank up the learning rate in DISCERN high enough, it produces language abnormalities that suggest schizophrenia."

After being re-trained with the elevated learning rate, DISCERN began putting itself at the center of fantastical, delusional stories that incorporated elements from other stories it had been told to recall. In one answer, for instance, DISCERN claimed responsibility for a terrorist bombing.

In another instance, DISCERN began showing evidence of"derailment"-replying to requests for a specific memory with a jumble of dissociated sentences, abrupt digressions and constant leaps from the first- to the third-person and back again.

"Information processing in neural networks tends to be like information processing in the human brain in many ways," says Grasemann."So the hope was that it would also break down in similar ways. And it did."

The parallel between their modified neural network and human schizophrenia isn't absolute proof the hyperlearning hypothesis is correct, says Grasemann. It is, however, support for the hypothesis, and also evidence of how useful neural networks can be in understanding the human brain.

"We have so much more control over neural networks than we could ever have over human subjects," he says."The hope is that this kind of modeling will help clinical research."

Source

Evolutionary Lessons for Wind Farm Efficiency

Senior Lecturer Dr Frank Neumann, from the School of Computer Science, is using a"selection of the fittest" step-by-step approach called"evolutionary algorithms" to optimise wind turbine placement. This takes into account wake effects, the minimum amount of land needed, wind factors and the complex aerodynamics of wind turbines.

"Renewable energy is playing an increasing role in the supply of energy worldwide and will help mitigate climate change," says Dr Neumann."To further increase the productivity of wind farms, we need to exploit methods that help to optimise their performance."

Dr Neumann says the question of exactly where wind turbines should be placed to gain maximum efficiency is highly complex."An evolutionary algorithm is a mathematical process where potential solutions keep being improved a step at a time until the optimum is reached," he says.

"You can think of it like parents producing a number of offspring, each with differing characteristics," he says."As with evolution, each population or 'set of solutions' from a new generation should get better. These solutions can be evaluated in parallel to speed up the computation."

Other biology-inspired algorithms to solve complex problems are based on ant colonies.

"Ant colony optimisation" uses the principle of ants finding the shortest way to a source of food from their nest.

"You can observe them in nature, they do it very efficiently communicating between each other using pheromone trails," says Dr Neumann."After a certain amount of time, they will have found the best route to the food -- problem solved. We can also solve human problems using the same principles through computer algorithms."

Dr Neumann has come to the University of Adelaide this year from Germany where he worked at the Max Planck Institute. He is working on wind turbine placement optimisation in collaboration with researchers at the Massachusetts Institute of Technology.

"Current approaches to solving this placement optimisation can only deal with a small number of turbines," Dr Neumann says."We have demonstrated an accurate and efficient algorithm for as many as 1000 turbines."

The researchers are now looking to fine-tune the algorithms even further using different models of wake effect and complex aerodynamic factors.

Source