Small data, Kalman filter, belief...

They're using Bayesian statistic. Probably added some expert belief to their prior distribution (informative prior). The whole article made it sound spicier than it has to be. The majority of the ML algorithms out there have too many parameters so it require a lot of data to overcome certain weaknesses like selection bias in tree ensemble, etc... So with small data it seems like ML are turning toward Bayesian. Also article may be missusing the term inference.