- Surprise-Driven Belief Revision 2007-04-10
Lorini, E., Castelfranchi, C. (2006); In Proceedings of Second Biennial Conference on Cognitive Science, St. Petersburg, 9-13 June, 2006.
- Using motor babbling and Hebb rules for modeling the development of reaching with obstacles and grasping 2008-05-15
Caligiore D., Ferrauto T., Parisi D., Accornero N., Capozza M., Baldassarre G. (2008). Using motor babbling and Hebb rules for modeling the development of reaching with obstacles and grasping. In Dillmann Rüdiger, Maloney Colette, Sandini Giulio, Asfour Tamim, Cheng Gordon, Metta Giorgio, Ude Ales (orgs.), International Conference on Cognitive Systems (CogSys2008) (In electronic format). (University of Karlsruhe, Karlsruhe, Germany, 2-4 April 2008)
- Brain waves encode rules for behavior 2012-11-21
- One of the biggest puzzles in neuroscience is how our brains encode thoughts, such as perceptions and memories, at the cellular level. Some evidence suggests that ensembles of neurons represent each unique piece of information, but no one knows just what these ensembles look like, or how they form.
A new study from researchers at MIT and Boston University (BU) sheds light on how neural ensembles form thoughts and support the flexibility to change one’s mind. The research team, led by Earl Miller, the Picower Professor of Neuroscience at MIT, identified groups of neurons that encode specific behavioral rules by oscillating in synchrony with each other.
The results suggest that the nature of conscious thought may be rhythmic, according to the researchers, who published their findings in the Nov. 21 issue of Neuron.
“As we talk, thoughts float in and out of our heads. Those are all ensembles forming and then reconfiguring to something else. It’s been a mystery how the brain does this,” says Miller, who is also a member of MIT’s Picower Institute for Learning and Memory. “That’s the fundamental problem that we’re talking about — the very nature of thought itself.”
Rules for behavior
The researchers identified two neural ensembles in the brains of monkeys trained to respond to objects based on either their color or orientation. This task requires cognitive flexibility — the ability to switch between two distinct sets of rules for behavior.
“Effectively what they’re doing is focusing on some parts of information in the world and ignoring others. Which behavior they’re doing depends on the context,” says Tim Buschman, an MIT postdoc and one of the lead authors of the paper.
As the animals switched between tasks, the researchers measured the brain waves produced in different locations throughout the prefrontal cortex, where most planning and thought takes place. Those waves are generated by rhythmic fluctuations of neurons’ electrical activity.
When the animals responded to objects based on orientation, the researchers found that certain neurons oscillated at high frequencies that produce so-called beta waves. When color was the required rule, a different ensemble of neurons oscillated in the beta frequency. Some neurons overlapped, belonging to more than one group, but each ensemble had its own distinctive pattern.
Interestingly, the researchers also saw oscillations in the low-frequency alpha range among neurons that make up the orientation rule ensemble, but only when the color rule was being applied. The researchers believe that the alpha waves, which have been associated with suppression of brain activity, help to quiet the neurons that trigger the orientation rule.
“What this suggests is that orientation was dominant, and color was weaker. The brain was throwing this blast of alpha at the orientation ensemble to shut it up, so the animal could use the weaker ensemble,” Miller says.
The findings could explain how the brain can create any appropriate behavioral response to the countless possible combinations of stimuli, rules and required actions, says Pascal Fries, director of the Ernst Strungmann Institute for Neuroscience in Frankfurt, Germany.
“We likely compose the appropriate neuronal assembly on the fly through synchronization,” says Fries, who was not part of the research team. “The number of combinatorial possibilities is enormous, just like the number of possible 10-digit telephone numbers is.”
Eric Denovellis, a graduate student at Boston University, is also a lead author of the paper. Other authors are Cinira Diogo, a former Picower Institute postdoc, and Daniel Bullock, a professor of cognitive and neural systems at BU.
Oscillation as consciousness
The researchers are now trying to figure out how these neural ensembles coordinate their activity as the brain switches back and forth between different rules, or thoughts. Some neuroscientists have theorized that deeper brain structures, such as the thalamus, handle this coordination, but no one knows for sure, Miller says. “It’s one of the biggest mysteries of cognition, what controls your thoughts,” he says.
This work could also help unravel the neural basis of consciousness.
“The most fundamental characteristic of consciousness is its limited capacity. You only can hold a very few thoughts in mind simultaneously,” Miller says. These oscillations may explain why that is: Previous studies have shown that when an animal is holding two thoughts in mind, two different ensembles oscillate in beta frequencies, out of phase with one another.
“That immediately suggests why there’s a limited capacity to consciousness: Only so many balls can be kept in the air at the same time, only a limited amount of information can fit into one oscillatory cycle,” Miller says. Disruptions of these oscillations may be involved in neurological disorders such as schizophrenia; studies have shown that patients with schizophrenia have reduced beta oscillations.
The research was funded by the National Science Foundation and the National Institute of Mental Health.
- The benefits of surprise in dynamic environments: from Theory to Pratice 2007-05-10
(submitted to) 2nd International Conference on Affective Computing and Intelligent Interaction – ACII-07, 2007.
- Metric State Space Reinforcement Learning for a Vision-Capable Mobile Robot 2007-04-17
V. Zhumatiy, F. Gomez, M. Hutter, and J. Schmidhuber: Metric State Space
Reinforcement Learning for a Vision-Capable Mobile Robot, Proc.
of the Int'l Conf. on Intelligent Autonomous Systems,
IAS-06, Tokyo, 2006.
- Shared Intention Revisited: The Limits of Egoism are Not the Limits of Individualism 2007-04-10
Tummolini, L. (submitted to Economics and Philosophy)
- The Interplay of Analogy-Making with Active Vision and Motor Control in Anticipatory Robots 2007-04-22
Kiryazov, K., Petkov, G., Grinberg, M. , Kokinov, B., and Balkenius, C. (2007). The Interplay of Analogy-Making with Active Vision and Motor Control in Anticipatory Robots. In Butz, M. et al. (Ed.) Anticipatory Behavior in Adaptive Learning Systems: From Brains to Individual and Social Behavior. LNAI, 4520, Springer-Verlag
- The robotic equivalent of a Swiss army knife 2012-11-30
The device doesn’t look like much: a caterpillar-sized assembly of metal rings and strips resembling something you might find buried in a home-workshop drawer. But the technology behind it, and the long-range possibilities it represents, are quite remarkable.
The little device is called a milli-motein — a name melding its millimeter-sized components and a motorized design inspired by proteins, which naturally fold themselves into incredibly complex shapes. This minuscule robot may be a harbinger of future devices that could fold themselves up into almost any shape imaginable.
The device was conceived by Neil Gershenfeld, head of MIT’s Center for Bits and Atoms, visiting scientist Ara Knaian and postdoctoral associate Kenneth Cheung, and is described in a paper presented recently at the 2012 Intelligent Robots and Systems conference. Its key feature, Gershenfeld says: “It’s effectively a one-dimensional robot that can be made in a continuous strip, without conventionally moving parts, and then folded into arbitrary shapes.”
To build the world’s smallest chain robot, the team had to invent an entirely new kind of motor: not only small and strong, but also able to hold its position firmly even with power switched off. The researchers met these needs with a new system called an electropermanent motor.
The motor is similar in principle to the giant electromagnets used in scrapyards to lift cars, in which a powerful permanent magnet (one that, like an ordinary bar magnet, requires no power) is paired with a weaker magnet (one whose magnetic field direction can be flipped by an electric current in a coil). The two magnets are designed so that their fields either add or cancel, depending on which way the switchable field points. Thus, the force of the powerful magnet can be turned off at will — such as to release a suspended car — without having to power an enormous electromagnet the whole time.
A four-segment milli-motein chain with a one-centimeter
In this new miniature version, a series of permanent magnets paired with electromagnets are arranged in a circle; they drive a steel ring that’s situated around them. The key innovation, Knaian explains, is that “they do not take power in either the on or the off state, but only use power in the changing state,” using minimal energy overall.
Photo: MIT Center for Bits and Atoms
The milli-motein concept follows up on a paper, published last year, which examined the theoretical possibility of assembling any desired 3-D shape simply by folding a long string of identical subunits. That paper, co-authored by Cheung, MIT professor Erik Demaine, alumnus Saul Griffith, and former Computer Science and Artificial Intelligence Laboratory research scientist Jonathan Bachrach, proved mathematically that it was possible for any 3-D shape to be reproduced by folding a sufficiently long string — and that it’s possible to figure out how to fold such a string, and the exact steps needed to successfully reach the desired endpoint.
“We showed that you could make such a universal system that’s very simple,” Cheung says. While he and his colleagues have not yet proved a way of always finding the optimal path to a given folded shape, they did find several useful strategies for arriving at practical folding sequences.
Demaine points out that the folding of the shape doesn’t have to be sequential, moving along the string one joint at a time. “Ideally, you’d like to do it all at once,” he says, with each of the joints folding themselves to the desired configuration simultaneously so that the loads are distributed.
Other researchers, including some at MIT, have explored the idea of fashioning reconfigurable robots from a batch of separate pieces that could self-assemble into different configurations — an approach sometimes called “programmable pebbles.” But Gershenfeld’s team found that a string of subunits capable of folding itself into any shape could be simpler in terms of control, power and communications than using separate pieces that must find each other and assemble in the right order. “You can just pass signals down the chain,” Knaian says.
It’s part of an overall approach, Gershenfeld explains, to “turning data into things.” In an article in the current issue of the magazine Foreign Affairs, he describes a technology roadmap for accomplishing that, and its policy implications. He and his colleagues have established a global network of more than 100 “fab labs” that provide community access to computer-controlled fabrication tools. Today, the design information is contained in an external computer rather than in the materials being manufactured, but the research goal is to digitize the materials themselves so that they can ultimately change their own shape, as the milli-motein does.
Hod Lipson, an associate professor of mechanical and aerospace engineering and computing and information science at Cornell University, says, "This result brings us closer to the idea of programmable matter — where computer programs and materials merge to form a new kind of matter whose shape and function can be programmed — not unlike biology. Many people are excited today to learn about 3-D printing and its ability to fabricate any shape; Gershenfeld’s group is already thinking about the next episode, where we don’t just control the shape of objects, but also their behavior."
The milli-motein is part of a family of such devices being explored at size scales ranging from protein-based “nanoassemblers” to a version where the chain is as big as a person, Gershenfeld says. Ultimately, a reconfigurable robot should be “small, cheap, durable and strong,” Knaian says, adding that right now, “it’s not possible to get all of those.” Still, he points out, “Biology is the existence proof that it is possible.”
The MIT researchers’ work could lead to robotic systems that can be dynamically reconfigured to do many different jobs rather than repeating a fixed function, and that can be produced much more cheaply than conventional robotics.
The development of the milli-motein included recent graduate Maxim Lobovsky SM '11 and undergraduate students Asa Oines and Peter Schmidt-Neilsen (who worked on the project as visiting high-school students). The work was supported by the U.S. Defense Advanced Research Projects Agency’s Maximum Mobility and Manipulation and Programmable Matter projects.