The pervasiveness of machine learning and big data algorithms is on the rise in today’s world. We see it applied everywhere, from predicting where you want to vacation next, to which brand of chocolate you’re going to buy. These impacts are more superficial, but there are other applications as well, applications that are solving problems that have so far been unsolved by us.
This piece covers a study of motor learning and coarticulation. Motor learning is the change, resulting usually from practice, in the capability for responding to external stimuli. Motor learning involves muscle memory and the recognition of oft-repeated patterns. This project is led by the dynamic Dr. S K M Varadhan of the Department of Applied Mechanics. Involving the creation of a new alphabet and typing system, as well as custom-made gloves for each participant, this is a truly novel project.
What if somebody told you that learning a particular skill will only need a certain number of days or hours of practice until you became decently competent at it? What if you knew exactly which exercises to practice in order to learn a skill in the minimum possible time? These are the questions being posed and simultaneously answered by this project. By inventing a new typing task and keyboard system, it is a systematic study of the time-evolution of people’s competency at a task that they are attempting for the first time. The trends observed will help greatly in the rehabilitation of those with motor neuron disorders, and will also facilitate the engineering of optimal and efficient learning curves, so to speak.
While it may seem like a trivial study, this is actually an investigation of how the central nervous system handles the nontrivial problem of learning a new task. Participants of the (currently ongoing) study will be shown words on a computer screen, and they will be required to type those out in real time using a specially designed, personally tailored glove keyboard (refer the image below). Various locations on the hand correspond to different letters of the alphabet. The glove is made as light and noninvasive as possible, because the unnatural nature of a peripheral on one’s hand may impede progress, as well as add an added layer of complexity to the learning.
The sensors attached to the glove are numerous and cutting-edge. They are electromagnetic sensors that can measure position with an accuracy of 1.27 microns. To put that in perspective, each hair on your head is between 20 and 200 microns thick. They also measure pitch, yaw and roll, yielding a six-dimensional vector of the sensor position at every instant of measurement. There are a total of 16 sensors on the glove, which translates to a huge amount of data being collected during the typing task.
Motor learning is ubiquitous. Ever-present in our lives, there are countless examples. Learning to ride a cycle, writing or typing out any language, all these are things that get better with practice. Slowly, we develop muscle memory and attain a proficiency that makes these tasks ‘second nature’. A pen or a guitar becomes an extension of our hand.
A key aspect to this proficiency that one attains is explained by the process of coarticulation. Coarticulation refers to the way the brain plaits together different vowels and consonants so that the final vocal output is a smooth whole. Coarticulation as a concept is not limited to speech, but can be extended to other situations too. For example, one coarticulates notes on a guitar, as well as keys while typing on a keyboard. When we know what comes next, we know exactly where to place our hands so as to make movements between discrete steps as fluid as possible.
The large amount of data collected causes a problem in computational power. For every fifteen minutes of typing, the size of data collected is of the order of many tens of megabytes. The study proceeds now with multiple participants, each typing for multiple sessions every day. This results in a massive data set, and anyone who’s ever worked on a machine learning problem knows that it is courting disaster to train a model on such a massive data set. This brings us to another aspect of this project, which aims to reduce the size of the data set collected. The sensors on a single hand record 96 different variables at each instant of time, which is huge. It is entirely possible, and almost always true, that some variables are redundant, and can be eliminated from the analysis altogether. This is a problem that has to be attacked from a machine learning perspective. What is interesting is that solving this problem will cause an advance not only in this project, but in other realms of modern computer science.
In scientific projects that are touted to make a big impact in the future, a common feature is the incredible dedication of the project members. While interviewing Dr. Varadhan, the project leader and his Masters and Doctoral students, the enthusiasm of the project members was overflowing. From the design of the experiments to the precise wooden contraption to capture the electromagnetic data from the sensors, every aspect has been taken care of with the utmost detail.
The applications of this contraption are many. As stated earlier, it will help in the rehabilitation of patients of motor neuron disorders. Finding the most optimal methods to learn new tasks, both ubiquitous and novel, will be greatly beneficial and will save time in a now time- and productivity-obsessed society. A study of the way we learn and coordinate muscles and neurons will add to our knowledge of human neuroscience. There may be windfalls in the way these data sets are handled in the future; in an age of interdisciplinary research, this project elegantly marries neuroscience and machine learning. This is a stepping stone to something more advanced, a marriage of contrasting fields that will only enrich both of them.
Meet the Prof
Dr. Varadhan S K M is an Assistant Professor in the Department of Applied Mechanics. He is a movement chauvinist. He loves analyzing movements of people. He believes in continuous learning (he still takes online courses). He loves interacting and connecting with students. When he’s not working, he loves playing with his daughter. You can know more about his research at https://home.iitm.ac.in/skm/index.html.
Meet the Author
Vishal Katariya is a fourth year B.Tech. student in Engineering Physics. He is interested in quantum physics among other things. He likes reading, writing, music and tennis, which includes a fanatical support of Roger Federer.
