Tag Archives: sense
#439400 A Neuron’s Sense of Timing Encodes ...
We like to think of brains as computers: A physical system that processes inputs and spits out outputs. But, obviously, what’s between your ears bears little resemblance to your laptop.
Computer scientists know the intimate details of how computers store and process information because they design and build them. But neuroscientists didn’t build brains, which makes them a bit like a piece of alien technology they’ve found and are trying to reverse engineer.
At this point, researchers have catalogued the components fairly well. We know the brain is a vast and intricate network of cells called neurons that communicate by way of electrical and chemical signals. What’s harder to figure out is how this network makes sense of the world.
To do that, scientists try to tie behavior to activity in the brain by listening to the chatter of its neurons firing. If neurons in a region get rowdy when a person is eating chocolate, well, those cells might be processing taste or directing chewing. This method has mostly focused on the frequency at which neurons fire—that is, how often they fire in a given period of time.
But frequency alone is an imprecise measure. For years, research in rats has suggested that when neurons fire relative to their peers—during navigation of spaces in particular—may also encode information. This process, in which the timing of some neurons grows increasingly out of step with their neighbors, is called “phase precession.”
It wasn’t known if phase precession was widespread in mammals, but recent studies have found it in bats and marmosets. And now, a new study has shown that it happens in humans too, strengthening the case that phase precession may occur across species.
The new study also found evidence of phase precession outside of spatial tasks, lending some weight to the idea it may be a more general process in learning throughout the brain.
The paper was published in the journal Cell last month by a Columbia University team of researchers led by neuroscientist and biomedical engineer Josh Jacobs.
The researchers say more studies are needed to flesh out the role of phase precession in the brain, and how or if it contributes to learning is still uncertain.
But to Salman Qasim, a post-doctoral fellow on Jacobs’ team and lead author of the paper, the patterns are tantalizing. “[Phase precession is] so prominent and prevalent in the rodent brain that it makes you want to assume it’s a generalizable mechanism,” he told Quanta Magazine this month.
Rat Brains to Human Brains
Though phase precession in rats has been studied for decades, it’s taken longer to unearth it in humans for a couple reasons. For one, it’s more challenging to study in humans at the level of neurons because it requires placing electrodes deep in the brain. Also, our patterns of brain activity are subtler and more complex, making them harder to untangle.
To solve the first challenge, the team analyzed decade-old recordings of neural chatter from 13 patients with drug-resistant epilepsy. As a part of their treatment, the patients had electrodes implanted to map the storms of activity during a seizure.
In one test, they navigated a two-dimensional virtual world—like a simple video game—on a laptop. Their brain activity was recorded as they were instructed to drive and drop off “passengers” at six stores around the perimeter of a rectangular track.
The team combed through this activity for hints of phase precession.
Active regions of the brain tend to fire together at a steady rate. These rhythms, called brain waves, are like a metronome or internal clock. Phase precession occurs when individual neurons fall out of step with the prevailing brain waves nearby. In navigation of spaces, like in this study, a particular type of neuron, called a “place cell,” fires earlier and earlier compared to its peers as the subject approaches and passes through a region. Its early firing eventually links up with the late firing of the next place cell in the chain, strengthening the synapse between the two and encoding a path through space.
In rats, theta waves in the hippocampus, which is a region associated with navigation, are strong and clear, making precession easier to pick out. In humans, they’re weaker and more variable. So the team used a clever statistical analysis to widen the observed wave frequencies into a range. And that’s when the phase precession clearly stood out.
This result lined up with prior navigation studies in rats. But the team went a step further.
In another part of the brain, the frontal cortex, they found phase precession in neurons not involved in navigation. The timing of these cells fell out of step with their neighbors as the subject achieved the goal of dropping passengers off at one of the stores. This indicated phase precession may also encode the sequence of steps leading up to a goal.
The findings, therefore, extend the occurrence of phase precession to humans and to new tasks and regions in the brain. The researchers say this suggests the phenomenon may be a general mechanism that encodes experiences over time. Indeed, other research—some very recent and not yet peer-reviewed—validates this idea, tying it to the processing of sounds, smells, and series of images.
And, the cherry on top, the process compresses experience to the length of a single brain wave. That is, an experience that takes seconds—say, a rat moving through several locations in the real world—is compressed to the fraction of a second it takes the associated neurons to fire in sequence.
In theory, this could help explain how we learn so fast from so few examples. Something artificial intelligence algorithms struggle to do.
As enticing as the research is, however, both the team involved in the study and other researchers say it’s still too early to draw definitive conclusions. There are other theories for how humans learn so quickly, and it’s possible phase precession is an artifact of the way the brain functions as opposed to a driver of its information processing.
That said, the results justify more serious investigation.
“Anyone who looks at brain activity as much as we do knows that it’s often a chaotic, stochastic mess,” Qasim told Wired last month. “So when you see some order emerge in that chaos, you want to ascribe to it some sort of functional purpose.”
Only time will tell if that order is a fundamental neural algorithm or something else.
Image Credit: Daniele Franchi / Unsplash Continue reading
Posted in Human Robots
#439243 Scientists Added a Sense of Touch to a ...
Most people probably underestimate how much our sense of touch helps us navigate the world around us. New research has made it crystal clear after a robotic arm with the ability to feel was able to halve the time it took for the user to complete tasks.
In recent years, rapid advances in both robotics and neural interfaces have brought the dream of bionic limbs (like the one sported by Luke Skywalker in the Star Wars movies) within touching distance. In 2019, researchers even unveiled a robotic prosthetic arm with a sense of touch that the user could control with their thoughts alone.
But so far, these devices have typically relied on connecting to nerves and muscles in the patient’s residual upper arm. That has meant the devices don’t work for those who have been paralyzed or whose injuries have caused too much damage to those tissues.
That may be about to change, though. For the first time, researchers have allowed a patient to control a robotic arm using a direct connection to their brain while simultaneously receiving sensory information from the device. And by closing the loop, the patient was able to complete tasks in half the time compared to controlling the arm without any feedback.
“The control is so intuitive that I’m basically just thinking about things as if I were moving my own arm,” patient Nathan Copeland, who has been working with researchers at the University of Pittsburgh for six years, told NPR.
The results, reported in Science, build on previous work from the same team that showed they could use implants in Copeland’s somatosensory cortex to trigger sensations localized to regions of his hand, despite him having lost feeling and control thanks to a spinal cord injury.
The 28-year-old had also previously controlled an external robotic arm using a neural interface wired up to his motor cortex, but in the latest experiment the researchers combined the two strands of research, with impressive results.
In a series of tasks designed to test dexterity, including moving objects of different shapes and sizes and pouring from one cup to another, Copeland was able to reduce the time he took to complete these tasks from a median of 20 seconds to just 10, and his performance was often equivalent to that of an able-bodied person.
The sensory information that Copeland receives from the arm is still fairly rudimentary. Sensors measure torque in the joints at the base of the robotic fingers, which is then translated into electrical signals and transmitted to the brain. He reported that the feedback didn’t feel natural, but more like pressure or a gentle tingling.
But that’s still a lot more information than cab be gleaned from simply watching the hand’s motions, which is all he had to go on before. And the approach required almost no training, unlike other popular approaches based on sensory substitution that stimulate a patch of skin or provide visual or audio cues that the patient has to learn to associate with tactile sensations.
“We still have a long way to go in terms of making the sensations more realistic and bringing this technology to people’s homes, but the closer we can get to recreating the normal inputs to the brain, the better off we will be,” Robert Gaunt, a co-author of the paper, said in a press release.
“When even limited and imperfect sensation is restored, the person’s performance improved in a pretty significant way.”
An external robotic arm is still a long way from a properly integrated prosthetic, and it will likely require significant work to squeeze all the required technology into a more portable package. But Bolu Ajiboye, a neural engineer from Case Western Reserve University, told Wired that providing realistic sensory signals directly to the brain, and in particular ones that are relayed in real time, is a significant advance.
In a related perspective in Science, Aldo Faisal of Imperial College London said that the integration of a sense of touch may not only boost the performance of prosthetics, but also give patients a greater sense of ownership over their replacement limbs.
The breakthrough, he added, also opens up a host of interesting lines of scientific inquiry, including whether similar approaches could help advance robotics or be used to augment human perception with non-biological sensors.
Image Credit: RAEng_Publications from Pixabay Continue reading
Posted in Human Robots
#439105 This Robot Taught Itself to Walk in a ...
Recently, in a Berkeley lab, a robot called Cassie taught itself to walk, a little like a toddler might. Through trial and error, it learned to move in a simulated world. Then its handlers sent it strolling through a minefield of real-world tests to see how it’d fare.
And, as it turns out, it fared pretty damn well. With no further fine-tuning, the robot—which is basically just a pair of legs—was able to walk in all directions, squat down while walking, right itself when pushed off balance, and adjust to different kinds of surfaces.
It’s the first time a machine learning approach known as reinforcement learning has been so successfully applied in two-legged robots.
This likely isn’t the first robot video you’ve seen, nor the most polished.
For years, the internet has been enthralled by videos of robots doing far more than walking and regaining their balance. All that is table stakes these days. Boston Dynamics, the heavyweight champ of robot videos, regularly releases mind-blowing footage of robots doing parkour, back flips, and complex dance routines. At times, it can seem the world of iRobot is just around the corner.
This sense of awe is well-earned. Boston Dynamics is one of the world’s top makers of advanced robots.
But they still have to meticulously hand program and choreograph the movements of the robots in their videos. This is a powerful approach, and the Boston Dynamics team has done incredible things with it.
In real-world situations, however, robots need to be robust and resilient. They need to regularly deal with the unexpected, and no amount of choreography will do. Which is how, it’s hoped, machine learning can help.
Reinforcement learning has been most famously exploited by Alphabet’s DeepMind to train algorithms that thrash humans at some the most difficult games. Simplistically, it’s modeled on the way we learn. Touch the stove, get burned, don’t touch the damn thing again; say please, get a jelly bean, politely ask for another.
In Cassie’s case, the Berkeley team used reinforcement learning to train an algorithm to walk in a simulation. It’s not the first AI to learn to walk in this manner. But going from simulation to the real world doesn’t always translate.
Subtle differences between the two can (literally) trip up a fledgling robot as it tries out its sim skills for the first time.
To overcome this challenge, the researchers used two simulations instead of one. The first simulation, an open source training environment called MuJoCo, was where the algorithm drew upon a large library of possible movements and, through trial and error, learned to apply them. The second simulation, called Matlab SimMechanics, served as a low-stakes testing ground that more precisely matched real-world conditions.
Once the algorithm was good enough, it graduated to Cassie.
And amazingly, it didn’t need further polishing. Said another way, when it was born into the physical world—it knew how to walk just fine. In addition, it was also quite robust. The researchers write that two motors in Cassie’s knee malfunctioned during the experiment, but the robot was able to adjust and keep on trucking.
Other labs have been hard at work applying machine learning to robotics.
Last year Google used reinforcement learning to train a (simpler) four-legged robot. And OpenAI has used it with robotic arms. Boston Dynamics, too, will likely explore ways to augment their robots with machine learning. New approaches—like this one aimed at training multi-skilled robots or this one offering continuous learning beyond training—may also move the dial. It’s early yet, however, and there’s no telling when machine learning will exceed more traditional methods.
And in the meantime, Boston Dynamics bots are testing the commercial waters.
Still, robotics researchers, who were not part of the Berkeley team, think the approach is promising. Edward Johns, head of Imperial College London’s Robot Learning Lab, told MIT Technology Review, “This is one of the most successful examples I have seen.”
The Berkeley team hopes to build on that success by trying out “more dynamic and agile behaviors.” So, might a self-taught parkour-Cassie be headed our way? We’ll see.
Image Credit: University of California Berkeley Hybrid Robotics via YouTube Continue reading
Posted in Human Robots