Gretchen Reynolds, Something in the Way We Move, NYTimes, Oct 23, 2019:
Each of us appears to have a unique way of moving, a physical “signature” that is ours alone, like our face or fingerprints, according to a remarkable new study of people and their muscles. The study, which used machine learning to find one-of-a-kind patterns in people’s muscular contractions, could have implications for our understanding of health, physical performance, personalized medicine and whether and why people can respond so differently to the same exercise.Intuitively, most of us probably know there is something in the way we move, and that that something defines us. In studies and daily life, most people can pick out their friends and loved ones, based solely on how they walk. At least one surveillance company also claims to be able to identify and track people using their gaits.But those identifications, whether fond or creepy, rely on external cues about how we look in motion and depend on anatomical features, such as height, limb length or how we swing our arms, which may not be stable. Wear heels, develop sore feet, limp, and you could move differently.Some scientists have speculated that other subtler, interior movement patterns, such as the ways in which our muscles fire in choreography with one another when we will our body to move, might be particular to each of us and relatively constant. But little research had delved into what our muscles are up to when we move.So, for the new study, which was published this month in the Journal of Applied Physiology, French and Australian researchers decided to turn to algorithms to ferret out whether unique, personal muscle patterns exist.
They recorded the muscle contractions from eight muscles on each of 80 volunteers as they rode stationary bikes and walked on treadmills in a human performance lab. They then fed the measurements to a machine-learning program in a supervised learning design. Once the system had been exposed to identified measurements it was then given unidentified measurements, which it was able to identify with 99 percent accuracy on the basis of readouts from eight muscles. Accuracy dropped to 80 percent with readouts from only two muscles.