My layman's understanding of earthquakes is that they have a heavy tailed probability distribution which makes predictions fundamentally difficult. Does ML (as in this approach) get around that somehow?